Upload
independent
View
0
Download
0
Embed Size (px)
Citation preview
1
Wagering in Australia: A retrospective behavioural analysis of
betting patterns based on player account data
Sally Gainsbury (a), Saalem Sadeque (b), Dick Mizerski (b), & Alex Blaszczynski (c)
(a) Centre for Gambling Education & Research, Southern Cross University
PO Box 157, Lismore NSW 2480 Australia
(b) Business School, University of Western Australia (M263)
35 Stirling Highway, Crawley WA 6009 Australia
(c) School of Psychology, The University of Sydney
Brennan MacCallum Building (A19), University of Sydney NSW 2006 Australia
This manuscript has been published in the Journal of Gambling Business & Economics.
Citation: Gainsbury, S., Sadeque, S., Mizerski, R., & Blaszczynski, A. (2012). Wagering in
Australia: A retrospective behavioural analysis of betting patterns based on player account
data. Journal of Gambling Business and Economics, 6(2), 50-68
The authors would like to acknowledge the European Association for the Study of Gambling who
provided funding for this research in the form of a Young Investigator Grant for the first author. Correspondence concerning this article should be addressed to: Sally Gainsbury, Centre for Gambling Education & Research, Southern Cross University, PO Box
157, Lismore NSW 2480, Australia. Email: [email protected].
2
Wagering in Australia: A retrospective behavioural analysis of
betting patterns based on player account data
Abstract
Gambling research often relies on self-report and cross-sectional data which is limited by inaccuracies
in recall. Analysis of behavioural data is necessary to advance conceptual understandings of gambling.
This paper analysed player account data of 11,394 customers of a large Australian wagering operator
over a ten-year period to investigate characteristics and betting patterns of account holders.
Comparisons were made between players based on the total number of bets placed. More frequent
bettors (those with greater total bet frequency counts), made smaller bets, but bet greater total amounts
and lost smaller proportions as compared to less frequent bettors. Less frequent bettors bet larger
single bets and lost a greater proportion of their total amounts bet. A minority of bettors accounted for
a disproportionately high number of bets but lost the lowest proportion of these. The results indicate
that players exhibit differential patterns of betting and subgroups of gamblers can be identified and
appropriately targeted with player education and responsible gambling strategies.
Keywords: gambling, betting, wagering, sports, behavioural analysis
3
Wagering in Australia: A retrospective behavioural analysis of
betting patterns based on player account data
Globally, sports and races fill an important role socially, economically and as a form of entertainment.
Betting or wagering is complementary to sporting events and a popular form of entertainment. For
example, on an annual basis 34% of adults in the UK bet on horse or dog race or sporting events
(Wardle et al., 2011), which does not include those who bet on non-sporting events and private
betting, which would increase participation rates, and similar rates are reported for Australia
(Productivity Commission, 2010). However, relatively little research has focused on this form of
gambling resulting in a limited understanding of consumer’s wagering behaviour. This paper aims to
identify patterns of wagering that differentiate subgroups of customers by analysing actual behavioural
data derived from a large sample of gamblers. Understanding common patterns of play will further the
understanding of how gambling products are typically used, which will aid the development of
appropriate consumer protection and customer interaction strategies..
Although race wagering has remained fairly stable over the past ten years in terms of participation and
expenditure, sports betting has increased driven by cable and digital television channels dedicated to a
wide range of events, increased Internet access enabling customers to bet online and follow statistics
and, liberalised regulation of wagering (Church-Sanders, 2011). Wagering, which includes betting on
both races and sports, is thought to be leading the global online gambling market, accounting for an
estimated 39% of online gambling, despite representing only 5% of total global gross gambling win
(ie, stakes less prizes) (Henwood, 2011). Increased marketing through sports sponsorship has
increased publicity and awareness of betting, and particularly Internet betting, in society (Lamont,
Hing, & Gainsbury, 2011) and anecdotal reports suggests that there has been an increase in young
males presenting for gambling treatment in relation to sports betting (University of Sydney, 2010).
Given the increased participation in online wagering (Gainsbury et al., 2012; Wardle et al., 2011),
4
consumers that engage in wagering represent an important target for research to further the
understanding of this gambling activity and the potential impact on individuals.
Methodological considerations in gambling research
A significant limitation of gambling research historically is its reliance on self-report measures of
gambling involvement (Blaszczynski, Ladouceur, Goulet, & Savard, 2006; Gainsbury, 2011;
Gainsbury & Blaszczynski, 2011; Hodgins & Makarchuk, 2003; McMillen & Wenzel, 2006; Walker,
2008; Williams & Volberg, 2009). Self-reported behaviours may be inaccurate as self-perception and
demand characteristics introduce response biases and reduce reliability and generalisability of findings
(Dickerson & Baron, 2000; Productivity Commission, 2010). Furthermore, self-reports of intended
behaviours by participants are often discordant with their actual behaviour (Baumeister, Vohs, &
Funder 2007). Most studies tend to view gamblers as a homogenous group and dismiss potential
implications of varied forms of gambling on etiological, personality or motivational factors
(Gainsbury, 2011; Holtgrave, 2009). This limits the extent to which research can contribute to an
understanding of gamblers based on different patterns of betting and consequentially devise
appropriate strategies to communicate with players and ensure play remains within appropriate limits.
The current study advances the field by comparing gamblers based on their betting frequency, which
will enable a greater understanding of different types of gamblers, driving further research.
Cross-sectional research designs are often used for cost and convenience purposes; however these do
not enable directions of causality to be determined. Longitudinal studies are costly, difficult to
implement, and often limited by attrition. Therefore, longitudinal studies that incorporate actual
behavioural and/or expenditure data based on player accounts would allow for more accurate
investigations of betting patterns and their variability over specified timeframes (Gainsbury, 2011;
Shaffer, Peller, LaPlante, Nelson, & LaBrie 2010). There is increasing evidence that suggests
electronically-recorded behavioural data (frequency, type, electronic interactions with operators)
obtained from gambling account holders can be used to effectively predict future behaviour and to
differentiate subgroups of gamblers (Braverman & Shaffer, 2010; Dragicevic, Tsogas, & Kudic, 2011;
5
Haefeli, Lischer, & Schwarx, 2011; Jolley, Mizerski & Olaru, 2006; Lam, 2006; LaBrie, LaPlante,
Nelson, Schumann, & Shaffer, 2007; LaBrie & Shaffer, 2010; Mizerski, Miller, Mizerski & Lam,
2004; Mizerski & Mizerski, 2001). These studies are based on recognition of the heterogeneity of
gamblers and their participation in gambling and the importance of understanding specific
involvement and risky behaviours to guide prevention and treatment initiatives. Analysis of
behavioural data is limited in that contextual variables, such as thoughts and emotions, are not
typically measured. However, it is argued that behavioural analysis has the advantage of overcoming
some limitations of self-report and cross-sectional studies and combined with multi-modal research
can be triangulated to evaluate conceptual models and theoretical postulates (Fantino, 2008;
Gainsbury, 2011).
Race wagering & Australian gambling
Australia represents a small but strong market for gambling and wagering. Australia had the highest
global adult per capita gambling expenditure; US$1,300 for every resident aged 17 and over as
compared to the US and Britain who spend less than US$400 per person on gambling (H2 Gambling
Capital as reported in The Economist). Betting on sports and races accounts for approximately 15% of
gambling losses (Australian Gaming Statistics as cited by Church-Sanders, 2011) and sports betting is
by far the fastest growing gambling segment, with a compound annual growth of 14.7% over the past
five years (IbisWorld, 2012).Bets may be placed in person with bookies at race tracks, at stand-alone
betting offices or agencies located in neighbourhood stores, hotels and clubs, or through telephone and
online accounts.
Wagering behaviour
Little research has been conducted examining behavioural characteristics and psychological profiles
associated with wagering. Woolley (2003) surveyed 2,945 account holders from two Australian
online wagering operators. The majority bet on races online at least once a week, although
participation in sports betting online appeared to be less frequent. There also appeared to be different
patterns of participation or subgroups of bettors, which is consistent with subsequent research (Lloyd
6
et al., 2010). Although this study provided useful insight into betting behaviour, the self-selected
sample may not be representative of all bettors and as the surveys were conducted almost ten years
ago, the results may not represent the current behaviour of online bettors. Furthermore, more recent
Australian research (Gainsbury et al., 2012) indicates that there may be differences between those who
gamble online and at land-based facilities, indicating that studies of online only betting may not
provide a comprehensive exploration of betting patterns.
Nonetheless, the majority of player behavioural data comes from online operators. In an examination
of more than 40,000 European online sports betting subscribers who opened accounts in 2005, LaBrie
et al. (2007) found players typically placed few small bets once a week and that only a small
percentage deviated from this pattern. Among those deviating was a subgroup displaying a pattern of
more frequent bets than the majority, but sustaining lower percentage losses. Subsequent studies with
the same cohort suggested there appeared to be a subgroup of gamblers characterised by potentially
less control over their play and more likely to repeatedly exceed pre-set individual limits on play
(Broda et al., 2008; LaPlante, Schumann, LaBrie, & Shaffer, 2008). Although these studies represent a
significant contribution to the field as they provide data on actual Internet gambling behaviour, they
are limited as they only provide information about a specific sample of bettors who opened accounts in
a defined period in 2005. The results do not provide information about players who have opened
accounts more recently and, as it is widely accepted that new customers to most businesses behave
significantly differently than already existing customers (Griffiths & Auer, 2011), this further limits
the extent to which results can be generalised to other populations.
In a Canadian population survey, Holtgrave (2009) found that gamblers who bet online and those who
bet on sports or horse races were distinctly different from gamblers who bet on other activities. These
types of gambling activities were associated with a greater likelihood that participants gambled
frequently, had greater gambling expenditure, and were more likely to have gambling-related
problems. Frequency of play has been found to be associated with and predictive of problem gambling
(Currie et al., 2006; Griffiths, Wardle, Orford, Sproston, & Erens, 2009; Hopley & Nicki, 2010; Lam
7
& Mizerski, 2009). An analysis of behavioural patterns related to potential gambling-related problems
indicated that players characterised by both high frequency of gambling and variability of bet sizes
during their first month of gambling were at higher risk of closing accounts due to gambling problems
(Braverman & Shaffer, 2010). However, although customers often closed accounts due to gambling-
related problems, no measures were used to verify that these individuals had gambling problems at a
clinical level. These results suggest that frequency of play is an important variable that may separate
subgroups of gamblers.
The current study
This study aimed to analyse comprehensive account-based data to investigate the characteristics,
betting patterns and behaviours of Australian wagerers. The objective was to investigate subgroups of
gamblers based on wagering frequency and to identify key betting characteristics and common
patterns of wagering. The purpose of the study was to undertake a comparative analysis of gambling
behaviours between more and less frequent bettors, determine the correlation between gambling
behaviours based on level of betting frequency, and identify factors distinguishing more and less
frequent bettors.
METHOD
Sample
An Australian wagering operator provided a database of de-identified account information for 22,849
individuals, with 11,394 customers making at least one bet. The percentage of opened accounts with
absent bets is similar to that reported in an online electronic gaming machine (EGM) study (Jolley et
al., 2006), and anecdotally with another online wagering provider (personal communication, requested
anonymity, 2011). The operator provides wagering services on thoroughbred, harness, and greyhounds
races, and sports events (including overseas events). Pari-mutuel betting (all bets accumulated in a
pool, operators subtract commissions and distribute balance to winning bettors in proportion to
wagered amounts) and fixed-odds betting products (where the odds are known at the time of the
8
placement of the wager) are offered with bets placed variably through; telephone, Internet, on-course
and networks of hotel, club and retail outlets. The website is one of the top ten Internet betting sites in
Australia (Alexa, cited by Church-Sanders, 2010).
The data period ranged from July 1, 2001 to November 24, 2010 and accounted for all bets placed (ie,
bets placed through all channels). The dataset included information on number of betting days, bets
placed, minimum and maximum bet value per single bet, total bet value for the specified timeframe,
and total dividend (ie, total amount returned to the player). All variables except for the first two were
expressed in Australian dollars (AUD$). The dataset also included the dates of the first bet and the last
bet made by each player.
Measures
Several measures were developed to enable data analysis; These include betting duration (difference
between first and last betting date), average dollar per bet (total bet amount divided by number of
bets), average number of bets per day (number of bets divided by number of betting days), frequency
of betting sessions (number of betting days divided by betting duration and expressed as a percentage),
net loss in dollars (difference between total amount bet and total dividends paid) and net loss
percentage (net loss divided by total amount bets).
Analysis
The dataset was divided into two groups using the median split based on the total number of bets
(frequency) made. Groups were labelled as more frequent (top 50% of the sample) and less frequent
bettors (lower 50% of the sample). Median value split has been widely used in academic literature (eg,
Bearden, Rose, & Teel, 1994; Gauri, Sudhir, & Talukdar, 2008; Greene & Greene, 2008; Hong &
Sternthal, 2010; Kilbourne & LaForge, 2010; Reinartz & Kumar, 2000; Schmittlein, Cooper, &
Morrison, 1993). The median split is particularly useful on strongly skewed distributions such as those
found in gambling (Jolley et al., 2006). The data exhibited the skewed Negative Binomial
Distributions (NBD) for variables under consideration. The frequency groups were created with the
9
aim of understanding the role of gambling variables for an actual gambling operator. In addition, a
discriminant analysis (DA) was conducted to identify the variables most useful in distinguishing
between the two frequency groups.
RESULTS
Demographics
Limited demographics information was provided for the account holders. Of the 10,655 customers that
provided personal details, 78% were male with a median age of 49 years (SD = 17.9 years). This data
was missing for 739 customers as it was not initially required to open an account in 2001; this has
become a requirement in subsequent years.
Player wagering behaviour
The average (mean) length of time that players actively placed wagers (duration) was 2,117.66 days
(SD = 1,304.34), which was highly variable over the ten year period under study. For the period in
which players had active accounts, they bet an average of 45.6 days and made an average of 717.6
bets. The mean percent betting days within the betting interval (frequency of betting sessions) was
6.93%, although the median was substantially lower at 1.56% with considerable variation in this
activity (SD = 16.64%).
As seen in Figure 1, the number of bets placed for the entire sample was highly skewed in the
direction of fewer bets. A majority of the players bet only a few times, and a small minority of bettors
accounted for the majority of bets. This is evidenced by the large standard deviation relative to means
and differences between mean and median values. This distribution reflects a pattern of behaviour that
fits the Negative Binomial Distribution (NBD) found in most frequently purchased products and
gambling (Ehrenberg, Uncles, & Goodhardt, 2004; Lam & Mizerski, 2009). Because of this type of
skewed distribution, player groups were created based on the median value for the total number of bets
(114). These groups were labelled as less frequent bettors (n=5,684) and more frequent bettors
(n=5,696). Table 1 provides the descriptive statistics for these two groups.
10
Insert Figure 1 & Table 1 about here
Analyses indicated that 91.78% of players lost money, on average players lost 34.07% of the total
amount wagered (SD = -29.94%, n = 11,069). Less and more frequent bettors had similar average
frequency of betting sessions during the period in which they actively placed bets with the operator
(see Table 1). However, less frequent bettors were active for a shorter duration but had a greater
average minimum bet value ($6.92 vs. $1.51), average dollars per bet ($18.82 vs. $14.05) and
percentage net loss (43.76% vs. 24.86%) than more frequent bettors.
To identify statistical differences between the two groups of involvement, Mann-Whitney U tests were
conducted on all the comparisons. The Mann-Whitney U test is preferable when assumptions of
normality cannot be met, as with the NBD (Allen & Bennett, 2010). In this test, the median values of
the variables are compared (Pallant, 2005). The analysis showed that the median values of all the
variables (see Table 1) between the more and less frequent bettors were significantly different
(P<.001). The more and less frequent bettor groups significantly differ in their patterns of gambling.
Those in the top 1% of the entire sample (n=113) based on total number of bets were classified as the
most frequent bettors. As shown in Table 1, the most frequent bettors had a lower average minimum
bet value, lower average dollars per bet, and lower percentage losses compared to the remainder of the
sample. The most frequent bettors accounted for 16.88% of the total bet value of the entire sample and
received 19.28% of total dividends paid to the entire sample. In comparison, the remainder of the
sample (n=11,267) accounted for 83.12% of the total bet value and received 80.72% of the
total dividends. The most frequent were combined with the more frequent bettors for further analysis.
Relationship between other gambling variables and number of bets
Correlation analyses were conducted to investigate specific associations between variables. As the
variables are skewed, Kendall’s tau-b (represented as ‘τ’) was used to examine the interrelationships
11
among the variables. Kendall’s tau-b is a suitable alternative when the assumptions of normality
cannot be met (Allen & Bennett, 2010). Kendall’s tends to provide better estimate of true correlation
in the population compared to Spearman’s Rho (Allen & Bennett, 2010; Malhotra, 2010), and
provides protection against Type I errors (Arndt, Turvey, & Andreasen., 1999). Moreover, Kendall’s
tau-b can be interpreted in the same way as Spearman’s Rho (Table 2).
Insert Table 2 about here
A number of significant relationships between variables were found, many of which make intuitive
sense given that groups were created based on betting frequency. As would be expected, the
correlation between number of betting days and number of bets was positive, and similar for both
groups. Similarly, the correlation between the total bet value and the total dividend was very high for
both groups. Of interest, the frequency of betting sessions was negatively correlated with duration
(how long the bettor has been with the provider) for both the groups. This effect was larger for less
frequent bettors (τ = -.48) than the more frequent bettors (τ = -.31).
Total bet value was positively correlated with percentage net loss. This effect was larger for the less (τ
= .27) as compared to the more frequent bettors (τ = .12) and again reflected the finding that less
frequent bettors tend to lose proportionally more. The number of betting days and number of bets was
also positively correlated with percentage net loss. This was larger for the less frequent bettors than
the more frequent bettors, indicating that the less frequent bettors lost a greater proportional amount of
bets. All correlations were significant at the 0.01 level, as indicated in Table 2.
Determining user group membership
The variables provided were analysed to try to determine group membership using multivariate
discriminant analysis (MDA). Only the original variables (number of betting days, duration, maximum
bet value, minimum bet value, and total bet value) were included as independent variables. Because
the new variables were a combination of the original variables, omitting them avoids the problem of
singularity which can severely affect the results of an MDA (Hair, Black, Babin, & Anderson, 2010;
Tabachnick & Fidell, 2007). The number of bets variable was also omitted because the groups were
12
created based on this variable. The total dividend was also omitted as this does not accurately measure
gambling behaviour, but it is based on external factors including other bettors. Table 1 shows that the
original variables are skewed, which is a normal occurrence in datasets such as this where a large
proportion of the sample wagers infrequently (LaBrie et al., 2007; LaBrie & Shaffer, 2010).
Accordingly, these variables were normalized through natural log transformation
An MDA requires a holdout sample for validating the discriminant function. Because of the large
number of cases (n = 11,380), the dataset was randomly divided into two groups. This resulted in
5,690 cases for the analysis sample which was used to generate the discriminant function. The other
group (holdout sample) was used to assess the internal validity of the discriminant function for
predicting the group membership. Since there is no prior theory to determine which of the five
variables considered in this analysis will be most useful in distinguishing between the user groups, a
step-wise MDA was employed. In a step-wise estimation method, independent variables are entered
into the discriminant function one at a time based on their discriminating power. Mahalanobis D2 was
employed to determine overall significance (Hair et al., 2010; Tabachnick & Fidell, 2007).
The MDA found a significant discriminant function (Wilks’ λ = 0.352; χ2 = 5936.73; df = 5, p<0.001)
in the model building sample. The canonical correlation was 0.81 which implied that the function
could explain 65.6% of the variance in the two user groups. The function could correctly classify
93.6% of analysis sample and this result was identical in the holdout sample. Table 3 shows the hit
ratios for the two user groups for both the samples. These hit ratios were compared against the
proportional chance (50%) and maximum chance criteria (50%). Hair et al. (2010) recommend that for
practical significance, the achieved classification accuracy must exceed the selected comparison
standard by at least 25 percent. The hit ratios for both the analysis and holdout samples exceeded this
accuracy (62.5%).
Insert Table 3 about here
13
The resultant MDA identified all five variables under consideration as being important in
distinguishing between the two player groups. Table 4 reports the mean values of log transformed
variables along with the discriminant loadings. Discriminant loadings are considered more useful for
interpretive purposes compared to discriminant coefficients (Hair et al., 2010; Tabachnick and Fidell
2007). Tabachnick and Fidell (2007) suggest that loadings over 0.33 can be considered eligible for
inclusion in the model. As can be seen from Table 4, only three (number of betting days, total bet
value and minimum bet value) out of the five variables qualify under this rule. An F-test of the
equality of group means supported these results (p<0.001). The results indicate that a greater number
of betting days and total bet value was predictive of being a more frequent bettor. A greater minimum
bet value was predictive of being a less frequent bettor.
Insert Table 4 about here
DISCUSSION
The large skew found in the gambling behaviour in the current betting sample is similar to the
observed and self-reported behaviour of gamblers in other studies (Jolley et al., 2006; LaBrie et al.,
2007; Lam & Mizerski, 2009; Mizerski et al., 2004; Shaffer et al. 2010) and shows that there are
distinct differences in how players engage with gambling activities. As betting frequency was seen to
differ greatly between players and has been used in previous research to group gamblers (Broda et al.,
2008; LaPlante et al., 2008), this variable was selected to create player groups for comparison. Less
frequent bettors typically bet higher minimum amounts and more dollars per bet than more frequent
bettors. However, the total amount bet and average number of bets was lower for less frequent than for
the more frequent bettors, indicating that over time, more frequent bettors tended to wager more
money. This may also indicate different betting strategies, whereby more frequent bettors spread their
investment over a larger number of bets to reduce risks, while less frequent bettors may be less
familiar with different types of bets or events to wager on.
14
As more frequent bettors lost a lower proportion of their wagers, this may indicate that this group are
more savvy bettors, picking odds and making bets in a more successful manner than less frequent
bettors, who may be less concerned with the outcome and spend less time conducting research to
inform their bets. These results are similar to those reported by Broda et al. (2008) and LaBrie et al.
(2007). This consistency between Australian and European bettors indicates that the difference
between bettors based on bet frequency may be widespread, although more research is needed to
confirm this.
Analysis of the most frequent bettors (top 1%) found that this group typically bet smaller amounts per
wager. Although they lost greater amounts overall, this was related to the greater total amounts bet and
closer analysis reveals that they lost a lower proportion of bets than the vast majority of players. This
indicates that although there is great variation amongst these players, many were very successful in
their betting. This again replicates the results found by La Brie et al. (2007) and indicates that there are
negative relationships between percent loss and aggregated total of money wagered and frequency of
betting. This group may represent semi-professional or professional wagerers who are knowledgeable
about betting and spend time researching the optimal bets to place and are willing to stake large sums
on the outcomes. As high frequency gamblers have also been targeted as a group of potential problem
gamblers (Broda et al., 2008; LaPlante et al., 2008), this subgroup of players clearly requires further
research to ensure that appropriate customer interaction and responsible gambling initiatives are
implemented with these players.
Across the whole sample, customers with shorter account duration (time over which the account was
active) tended to make bets less frequently than those with a longer active betting history. This was
particularly pronounced for the least frequent bettors, suggesting that customers may have been trying
out or testing the operator to see if they enjoyed the activity, but did not overly enjoy betting or betting
with the operator. Given the large number of operators that offer gambling and the relative ease of
opening an account it is not surprising that some customer may test out wagering with an operator and
15
fail to continue. These customers may be of particular interest to online operators as a target for
retention strategies.
The discriminant analysis results showed that the key variables that distinguish between the less and
more frequent betting groups are the number of betting days, total bet value, and minimum bet value.
Specifically, players that place a greater total number of bets appear to place bets on a greater number
of days and account for large dollar amounts. In contrast, bettors who placed a lower total number of
bets participate in wagering less often but on average made higher bets. Future research may
investigate how wagering providers cater to different customers and whether they tend to may make
greater profits from low frequency, but higher bet average customers, who may lose a greater
proportion of their wagers. it is also important to investigate the extent to which these players also bet
on other gambling activities as negative financial consequences of gambling may place them at risk of
gambling-related harms.
The current study found that although more frequent betting was associated with large losses, the
proportion of losses to amounts wagered was lower for more frequent bettors than for less frequent
bettors. This questions the use of the variable of betting frequency to classify bettors as potential
problem gamblers, although the current study did not include any contextual variables such as
discretionary disposable income, psychological wellbeing and impact of gambling. Therefore,
although frequency of betting involvement may be a useful variable to classify and analyse bettors,
more research is needed to investigate whether this is an appropriate variable for the identification of
problem gamblers.
Although the implications of frequency of betting are not clearly defined this is an important variable
to use for the classification of gamblers. Based on an analysis of survey data of Canadian gamblers,
Humphreys et al. (2011) argued that frequency of participation is a better measure of the intensity of
consumer participation in gambling than reported expenditure. Self-reported expenditure data is often
inaccurate due to respondent’s varied understanding of expenditure in addition to recall bias
16
(Gainsbury et al., 2012; Wood & Williams, 2007). Self-report of gambling frequency may be more
accurate as this is not as complex to calculate as expenditure and therefore, this variable may be more
accurate in predicting future gambling behaviour. Despite the limitations of self-report data, this
method has unique advantages as it is able to examine more personal and subjective variables, such as
psychological characteristics. Therefore, future research may benefit by taking a multimodal approach
and considering both behavioural and self-report data to further the understanding of consumer
gambling behaviour.
As with most gambling research, this study is not without its limitations, some of which have already
been mentioned. Similar to other behavioural analyses of Internet gamblers, this study is limited to the
betting behaviour of customers to one unique gambling operator and may not be representative of
customers of other gambling operators. However, a key strength is the inclusion of customers that bet
via multiple channels, as opposed to only online gamblers. Few contextual variables were available
and no cognitive or emotive variables were measured, making it impossible to identify problem
gamblers; to do so require validation using clinical interviews or problem gambling measures. Future
research should attempt to overcome these limitations by combining player account behavioural
analysis with self-report data and scales. Comparisons should also be made between player account
data provided by various wagering operators to investigate if there are different betting behavioural
patterns found between gambling providers. In the current study, bet frequency was used for analytical
comparison given its identification being associated with problem gambling in previous studies
(Currie et al., 2006; Griffiths, Wardle, Orford, Sproston, & Erens, 2009; Hopley & Nicki, 2010;
LaBrie & Shaffer, 2011; Lam & Mizerski, 2009). The number of bets placed does not necessarily
reflect a high degree of riskiness or involvement given the importance of factors such as the odds of
winning. Future research should explore betting behaviour in an exploratory manner to investigate
what other variables may separate groups of bettors. Furthermore, ongoing research should consider
other variables relevant to betting involvement in examining subgroups of bettors, including
expenditure, bet size and the number of betting days.
17
Despite these limitations, this study has many important strengths; this is the first behavioural analysis
of actual wagering behaviour in Australia and provides insight into how customers place bets over a
ten year period. Previous studies have used a more limited time period and ongoing research will
further examine this longitudinal dataset. The finding of distinctive characteristics in betting patterns
based on gambling frequency has important implications for public policy, industry, research and
treatment providers. Recently, a federal parliamentary committee in Australia recommended the
introduction of ‘low intensity’ poker machines with a maximum of $1 bet per game (Productivity
Commission, 2010). Although our results are in the context of wagering, less frequent bettors
accounted for higher average bet value per bet than the more frequent bettors. Many of these less
frequent bettors could potentially be light and occasional (ie, Melbourne Cup Horse race, Football
Finals) recreational gamblers. This may suggest that the proposed change in maximum bet value may
affect the less frequent bettors rather than more frequent bettors, which may not be an optimal
responsible gambling strategy given previous research indicating that more frequent gamblers are at
greater risk of harm (Currie et al., 2006; Griffiths, Wardle, Orford, Sproston, & Erens, 2009; Hopley
& Nicki, 2010; Lam & Mizerski, 2009).
This study indicates that there are significant differences between groups of bettors based on their
betting patterns. Bettors that placed a greater number of bets lost more money overall, but were more
successful than less frequent bettors in terms of the proportion of wagers lost. This may indicate that
more frequent bettors are more skilled, or put more time and effort into placing smaller, but more
frequent bets to minimise losses as compared to less frequent bettors who place larger single bets and
tend to be less successful. These results have implications for betting operators in terms of their
approach to customers based on betting behaviour. Policy makers may consider initiatives to target
less frequent, recreational players, to ensure that these individuals spend within their means and ensure
the activity does not cause problems, as well as more frequent bettors who still lose greater overall
amounts. Future research should analyse behavioural data to further explore the development of
patterns of play and potentially identify players that may be at risk of experiencing gambling-related
harms.
19
REFERENCES
P Allen and K Bennett PASW statistics by SPSS: A practical guide, version 18.0 (South
Melbourne, Victoria, Cengage Learning Australia Pty Limited, 1st edition, 2010).
S Arndt, C Turvey and NC Andreasen. ‘Correlating and predicting psychiatric symptom
ratings: Spearman’s r versus Kendall’s tau correlation’ (1999) Journal of Psychiatric
Research, 33, 97.
RF Baumeister, KD Vohs and DC Funder ‘Psychology as the science of self-reports and
finger movements: Whatever happened to actual behavior?’ (2007) Perspectives on
Psychological Science, 2 , 396.
W Bearden, RL Rose and JE Teel ‘Correlates of conformity in the consumption of illicit
drugs and alcohol’ (1994) Journal of Business Research, 30, 25.
A Blaszczynski, R Ladouceur, A Goulet and C Savard ‘How much do you spend gambling?
Ambiguities in questionnaire items assessing expenditure’ (2006) International
Gambling Studies, 6, 123.
J Braverman and HJ Shaffer ‘How do gamblers start gambling: Identifying behavioural
markers for high-risk internet gambling.’ (2010). European Journal of Public Health.
Advance online publication. doi: 10.1093/eurpub/ckp232
A Broda, D LaPlante, SE Nelson, RA LaBrie, LB Bosworth and HJ Shaffer ‘Virtual harm
reduction efforts for Internet gambling: Effects of deposit limits on actual Internet
sports gambling behavior’ (2008) Harm Reduction Journal, 5, 27.
R Church-Sanders The global business of online casinos: Outlook, forecasts and analysis
(London, iGaming Business , 2010).
R Church-Sanders Online sports betting: A market assessment and outlook (London:,
iGaming Business, 2011).
20
SR Currie, DC Hodgins, J Wang, N el-Guebaly, H Wynne and S Chen ‘Risk of harm among
gamblers in the general population as a function of level or participation in gambling
activities’ (2006) Addiction, 101, 570.
M Dickerson and E Baron ‘Contemporary issues and future directions for research into
pathological gambling’ (2000) Addiction, 95, 1145.
S Dragicevic, G Tsogas and A Kudic ‘Analysis of casino online gambling data in relation to
behavioural risk markers for high-risk gambling and player protection’ (2011)
International Gambling Studies, 11, 377.
SC Ehrenberg, M Uncles and G Goodhardt ‘Understanding brand performance measures:
Using Dirichlet benchmarks’ (2004) Journal of Business Research, 57, 1307.
E Fantino ‘Behavior analysis: Thriving, but how about its future?’ (2008) Journal of
Experimental Analysis of Behavior, 89, 125.
S Gainsbury ‘Player Account-based Gambling: Potentials for Behaviour-based Research
Methodologies’ (2011) International Gambling Studies, 11, 153.
S Gainsbury and A Blaszczynski ‘A systematic review of Internet-based therapy for the
treatment of addictions’ (2011) Clinical Psychology Review, 31, 490.
S Gainsbury, R Wood, A Russell, N Hing and A Blaszczynski. ‘A digital revolution:
Comparison of demographic profiles, attitudes and gambling behaviour of Internet and
non-Internet gamblers’ (2012) Computers in Human Behaviour, 28(4), 1388.
DK Gauri, K Sudhir and D Talukdar ‘The temporal and spatial dimensions of price search:
Insights from matching household survey and purchase data’ (2008) Journal of
Marketing Research, 45, 226.
H Greene and S Greene ‘Enhancing segmentation systems’ (2008) Journal of Targeting,
Measurement, and Analysis for Marketing, 16, 298.
21
M Griffiths and M Auer ‘Approaches to understanding online versus offline gambling
impacts’ (2011) Casino & Gaming International, 3, 45.
M Griffiths, H Wardle, J Orford, K Sproston and B Erens ‘Internet gambling, health,
smoking and alcohol use: Findings from the 2007 British Gambling Prevalence
Survey’ (2009) International Journal of Mental Health and Addictions, 9, 1.
J Haefeli, S Lischer and J Schwarx ‘Early detection items and responsible gambling features
for online gambling’ (2011) International Gambling Studies, 11, 273.
JF Hair, WC Black, BJ Babin and RE Anderson Multivariate data analysis: A global
perspective . (New Jersey, Pearson Prentice Hall, ).. 7th
edition, 2010).
D Henwood ‘Sports betting: The big picture’ (2011) Presentation at EGR Live, Business
Design Centre, London. May 3-4, 2011. Retrieved from:
http://www.h2gc.com/news.php?article=H2+Gambling+Capital+Presentations+-
+EGR+Live+May+2011
DC Hodgins and K Makarchuk ‘Trusting problem gamblers: Reliability and validity of self-
reported gambling behavior’ (2003) Psychology of Addictive Behaviors, 17, 244.
T Holtgraves ‘Gambling, gambling activities, and problem gambling’ (2009) Psychology of
Addictive Behavior, 23, 295.
J Hong and B Sternthal ‘The effects of consumer prior knowledge and processing strategies
on judgments’ (2010) Journal of Marketing Research, 47, 301.
AAB Hopley and RM Nicki ‘Predictive factors of excessive online poker playing’ (2010)
Cyberpsychology, Behavior and Social Networking, 13, 379.
BR Humphreys, YS Lee and BP Soebbing ‘Modelling consumers’ participation in gambling
markets and frequency of gambling’ (2011) Journal of Gambling Business and
Economics, 5, 1.
22
IBISWorld ‘Footy season kicks off the annual serge in sports betting’ (2012, March 21)
IBISWorld, Melbourne.
B Jolley, R Mizerski and D Olaru ‘How habit and satisfaction affects player retention for
online gambling’ (2006) Journal of Business Research, 59, 770.
WE Kilbourne and MC LaForge ‘Materialism and its relationship to individual values’ (2010)
Psychology and Marketing, 27, 780.
RA LaBrie, DA LaPlante, SE Nelson, A Schumann and HJ Shaffer ‘Assessing the playing
field: A prospective longitudinal study of Internet sports gambling behavior’ (2007)
Journal of Gambling Studies, 23, 347.
R LaBrie and HJ Shaffer ‘Identifying behavioural markers of disordered Internet sports
gambling’ (2010) Addiction Research & Theory, 19, 56.
D Lam ‘Applicability of the duplication of purchase law to gaming’ (2006) UNLV Gaming
Research & Review Journal, 10, 55.
D Lam, D and R Mizerski ‘An investigation into gambling purchases using the NBD and
NBD-Dirichlet models’ (2009) Marketing Letters, 20, 263.
M Lamont, N Hing and S Gainsbury ‘Gambling on sport sponsorship: A conceptual
framework for research and regulatory review’ (2011) Sport Management Review, 14,
246.
DA LaPlante, A Schumann, RA LaBrieand HJ Shaffer ‘Population trends in Internet sports
gambling’ (2008) Computers in Human Behavior, 24, 2399.
J Lloyd, H Doll, K Hawton, WH Dutton, JR Geddes, GM Goodwin and RD Rogers ‘Internet
gamblers: a latent class analysis of their behaviours and health experiences’ (2010)
Journal of Gambling Studies 26, 387.
NK Malhotra Marketing research: An applied orientation (New Jersey, Prentice-Hall, 6th
edition, 2010).
23
J McMillen and M Wenzel ‘Measuring problem gambling: Assessment of three prevalence
screens’ (2006) International Gambling Studies, 6, 147.
D Mizerski, R Miller, K Mizerski and D Lam The stochastic nature of purchasing a state’s
lottery products’ (2004) Australasian Marketing Journal, 12, 56.
D Mizerski K Mizerski ‘The effect and implications for a stochastic pattern of lotto game
play’ (2001) International Gambling Studies, 1, 132.
J Pallant SPSS survival manual (Berkshire, United Kingdom, Open University Press, 2nd
edition, 2005).
Productivity Commission Gambling (Canberra, Report no. 50, 2010).
WJ Reinartz and V Kumar ‘On profitability of long-life customers in a noncontractual setting:
An empirical investigation and implications for marketing’ (2000) Journal of
Marketing, 64, 17.
Roy Morgan Research Roy Morgan Gambling Monitor (Sydney, Roy Morgan Research,
2011).
D Schmittlein, LG Cooper and D Morrison ‘Truth in the concentration in the land of (80/20)
laws’ (1993) Marketing Science, 12, 167.
HJ Shaffer, AJ Peller, DA LaPlante, SE Nelson and RA LaBrie ‘Toward a paradigm shift in
internet gambling research: From opinion and self-report to actual behaviour’ (2010)
Addiction Research and Theory, 18, 270.
BG Tabachnick and LS Fidell Using multivariate statistics (New Jersey, Pearson Education,
5th
edition, 2007).
The Economist online ‘The biggest losers’ (2011, May 16) The Economist Online. Retrieved
from http://www.economist.com/blogs/dailychart/2011/05/gambling
University of Sydney ‘Internet adds to gambling woes’ (2011, Mar 14) Science Alert.
Retrieved from: http://www.sciencealert.com.au/news/20111503-21947.html
24
DM Walker Challenges that confront researchers on estimating the social costs of gambling
(Washington, DC, American Gaming Association, 2008).
H Wardle, A Moody, S Spence, J Orford, R Volberg, D Jotangia, M Griffiths, D Hussey and
F Dobbie ‘British gambling prevalence survey 2010’ (London, National Centre for
Social Research, 2011).
R Williams and RA Volberg ‘Impact of survey description, administration format, and
exclusionary criteria on population prevalence rates of problem gambling’ (2009)
International Gambling Studies, 9, 101.
R Woolley ‘Mapping internet gambling: Emerging modes of online participation in wagering
and sports betting’ (2003) International Gambling Studies, 3, 3.
R Wood and R Williams ‘How much money do you spend on gambling? The comparative
validity of questions wordings used to assess gambling expenditure’ (2007)
International Journal of Social Research Methodology, 10, 63.
25
Table 1: Descriptive statistics of less frequent bettors vs. more frequent bettors vs. most
frequent bettors group (frequency = number of bets)
Less Frequent Bettors
(n=5,684) Bottom 50% of the full
sample
More Frequent Bettors
(n=5,696) Top 50% of the full
sample
Most Frequent Bettors (n=
113) Top 1% of the full sample
Variable Mean (SD) Median Mean(SD) Median Mean(SD) Median No. of betting
days* 9.29 (10.18) 6 81.94 (72.71) 57 252.32
(78.76) 263
No. of bets
(frequency)* 36.78
(32.08) 27 1,398.73
(2,968.31) 487.5 17,273.48
(8,765.96) 14,079
Minimum bet
value ($) 6.918 (24.61) 3 1.51 (2.26) 1 .57 (.21) .50
Maximum bet
value ($) 65.12
(459.19) 20 133.60
(551.43) 45 374.22
(1,749.63) 43
Total bet value
($)* 619.86
(2,145.83) 206 15,322.84
(84,831.17) 2,748.07 135,632.35
(370,000) 48,123
Total Dividend
($)* 370.47
(1,118.05) 99.50 11,973.89
(71,709.48) 2,748.08 119,340.44
(354,000) 37,846.61
Duration (days)** 1,915.75
(1,327.19) 2,234.5 2,319.63
(1,249.04) 3,073.5 2,519.03
(1,191.91) 3,336
Average number
of bets (based on
total betting days)
5.46 (5.99) 3.64 14.89 (17.06) 10.27 77.64
(53.48) 62.38
Average dollar per
bet ($) 18.82
(63.98) 8.46 14.05
(41.30) 6.69 7.54 (20.80) 3.15
Frequency of
betting sessions
(% days during
active
membership)
6.91 (20.90) .44 6.96 (10.81) 3.15 17.60
(19.16) 9.86
Net Loss ($)* 190.62
(825.57) 76.33 3,182.72
(26,685.11) 833.22 15,571.46
(69,496.64) 7,419.90
Net Loss (%)* 43.76 (75.63) 48.24 24.86 (21.66) 23.48 19.56 (9.14) 20.71
*Totals are for the entire period between 01.07.01 and 24.11.10; ** Number of days between
individuals’ first and last bets (within the period from 01.07.01 and 24.11.10)
26
Table 2: Correlations among gambling behaviour variables for less frequent bettors (above
diagonal) and more frequent bettors (below diagonal)
Variable
No. of
betting days
No.
of bets
Minimu
m bet value ($)
Maximu
m bet value ($)
Total bet
value
($)
Total
Dividend ($)
Duratio
n (days)
Average
number
of bets
Average dollar
per bet
($)
Frequenc
y of visit (%)
Net
Loss ($)
Net
Loss (%)
No. of betting
days -
.6
0 -.13 .24 .54 .48 .15 -.09 .08 .39 -.32 .25
No. of
bets .56 - -.29 .20 .59 .55 .09 .34 -.03 .28 -.36 .28
Minimu
m bet
value ($) -.07
-
.2
6 - .22
.02
* .01
ns .02* -.32 .49 -.08
.00n
s -
.02ns
Maximum bet
value ($) .08
.0
7 .12 - .56 .42 .03 .00
ns .65 .13 -.41 .11
Total bet
value ($) .43 .5
2 -.01
ns .47 - .71 .08 .17 .38 .25 -.54 .27
Total
Dividend ($)
.43 .5
2 -.01
ns .43 .88 - .08 .17 .29 .23 -.23 .56
Duration (days) .24
.1
5 .04 .01
ns .12 .13 - -.06 .02* -.48 -.03 .06
Average
number
of bets
-
.01ns
.4
3 -.33 -.00
ns .25 .26 -.05 - -.17 .00
ns -.12 .10
Average dollar per
bet ($) -.04
-
.1
2 .31 .62 .36 .33 -.01
ns -.16 - .04 -.28 .06
Frequenc
y of betting
sessions
(%)
.46 .3
3 -.1 .03 .24 .24 -.31 .07 -.06 - -.17 .11
Net Loss
($) -.35 -
.4
3 .07 -.38 -.65 -.53 -.09 -.23 -.26 -.20 - .19
Net Loss (%) .09
.0
9 .07 .03 .12 .24 .04 .06 .04 .05 .23 -
Note: Nonparametric Kendall’s tau-b; all correlation is significant at 0.01 level unless marked
with * and ns
.
*Correlation is significant at 0.05 level
ns
Correlation is non-significant
27
Table 3: Classification results for both analysis and holdout samples
Predicted group membership (%) Sample size (n)
Actual group (%) Less
frequent bettors
More frequent bettors
Analysis sample
Less frequent bettors
91.7 8.3 2,859
More frequent bettors
4.6 95.4 2,831
Holdout sample
Less frequent bettors
92.1 7.9 2,825
More frequent bettors
4.9 95.1 2,865
Note: 93.6 percentage of analysis and holdout sample correctly classified.
Table 4: Summary of discriminant analysis
Analysis sample (n=5,690) Holdout sample (n=5,690)
Variables Log Mean Discriminant loading
Log Mean Discriminant loading
Less
frequent bettors (n=2,859)
More
frequent bettors (n=2,831)
Less
frequent bettors (n=2,825)
More
frequent bettors (n=2,865)
Number of betting days
1.69 4.05 0.87 1.69 4.02 0.88
Total bet value ($)
.10 .13 0.76 .10 .13 0.79
Minimum bet
value ($) .06 .05 -0.41 .06 .05 -0.38
Maximum bet
value ($) .08 .08 0.20 .08 .08 0.22
Duration (days) 6.74 7.43 0.16 6.76 7.39 0.15
Note: Group centroids (analysis sample): Less frequent bettors = -1.35; More frequent bettors = 1.36
Group centroids (holdout sample): Less frequent bettors = -1.35; More frequent bettors = 1.33