Background: Participation in Internet gambling is growing rapidly, as is concern about its possible effects on the public's health. This article reports the results of the first prospective longitudinal study of actual Internet casino gambling behaviour. Methods: Data include 2 years of recorded Internet betting activity by a cohort of gamblers who subscribed to an Internet gambling service during February 2005. We examined computer records of each transaction and transformed them into measures of gambling involvement. The sample included 4222 gamblers who played casino games. Results: The median betting behaviour was to play casino games once every 2 weeks during a period of 9 months. Subscribers placed a median of 49 bets of €4 each playing day. Subscribers lost a median of 5.5% of total monies wagered. We determined a group of heavily involved bettors whose activity exceeded that of 95% of the sample; these players bet every fifth day during 17.5 months. On each playing day, these most involved bettors placed a median of 188 bets of €25. Their median percent of wagers lost, 2.5%, was smaller than that lost by the total sample. Conclusion: Our findings suggest that Internet casino betting behaviour results in modest costs for most players, while some, roughly 5%, have larger losses. The findings also show the need to consider time spent as a marker of disordered gambling. These findings provide the evidence to steer public health debates away from speculation and toward the creation of empirically-based strategies to protect the public health.
Pathological gambling is a public health problem associated with many physiological, psychological and social repercussions—some of which are shared with other expressions of addiction and some of which are unique to excessive gambling behaviour.1 There are many influences on the prevalence of pathological gambling among the general population.2,3 For example, researchers have identified an association between increases in opportunities to gamble and changes in population-level gambling behaviour.4,5 Researchers and gambling advocates alike worry that Internet-related increases in gambling opportunities can lead to excessive gambling among segments of the population.6–12 To date, very little research has examined population-level Internet gambling behaviour. This article provides the first glimpse into the actual gambling behaviour, as opposed to self-reported gambling behaviour, of online casino gamblers.
Internet gambling and public health
Rates of Internet gambling currently are low compared to other types of gambling.13–16 However, researchers and advocates have suggested that online gambling will grow and attract players because of some uniquely appealing aspects, such as anonymity, proximity and a greater sense of control.17–21 Moreover, the loose regulations implemented by many gambling websites,22 combined with younger generations’ familiarity with, interest in and access to computer technology, has created uneasiness concerning the possibility of an increase in underage gambling.17
Studies have reported higher prevalence rates of Internet gambling in special populations, suggesting differential popularity and potency of Internet gambling for these groups.15,23–25 For example, Ladd and Petry26 investigated the gambling behaviour of people seeking treatment at University of Connecticut health clinics (n = 389) and found that 8.1% of the cohort affirmed gambling on the Internet in their lifetime; those who gambled on the Internet were more likely to be younger, non-Caucasian and have higher scores on the South Oaks Gambling Screen27 in comparison to non-Internet gamblers. However, these prevalence studies have gathered data via self-report measures, which often provide inaccurate or biased data due to participants’ emotional responses to the events in question and difficulty with recall.28 Baumeister et al.21 simply state that people's accounts of their actions do not often correlate with their actual behaviours and thus these discrepancies lead to inaccurate reports. Evaluating gamblers’ actual online behaviour provides a valid record of gambling activities and more accurate knowledge about players’ behaviour patterns.
Assessing actual online gambling behaviour
LaBrie et al.,29 provided the first report of actual gambling behaviours of a longitudinal sample of Internet sports gamblers. This study revealed primarily moderate gambling behaviour at the population level (i.e. 2.5 bets of €4 each every fourth day), suggesting that Internet sports gambling does not encourage excessive gambling for many players. However, the research also showed that a small percentage of subscribers (i.e. 1%) exhibited behaviours that deviated markedly from the norm (i.e. the median activity profile of this group was to place 4.7 bets of €44 each every other day).
Similarly, analyses of temporal patterns of sports gambling behaviour showed evidence of rapid population-level adaptation to online gambling.30 The observed cycle of sports gambling activity indicated short-term increases in activity followed by a swift decline, a model consistent with prototypical public health adaptation curves.12 The rapid adaptation might be a result of previous gambling experience; the novelty of the Internet might have generated the initial short-term increases in activity. LaPlante and colleagues30 noted a small segment of the population (1%) which did not adapt, consistent with the findings of LaBrie et al.29
Our previous research29 provided a unique description of actual Internet sports gambling, and a description of actual Internet poker play is forthcoming. Both types of betting are referred to as ‘skill’ games. In comparison, chance governs casino games entirely. As a result, we expect that casino game play will differ in a few important ways from sports gambling and poker play. Structurally, casino play is more rapid; therefore, we expect that volume measures, such as number of bets, will be greater for casino gambling than for sports betting. Our research focusing on the gambling behaviour of 2356 people in treatment for gambling-related problems31–33 suggests that casino games are the game of choice for people seeking treatment. Despite the absence of epidemiologic evidence other than that presented here, we hypothesize that individuals betting in virtual casinos will exhibit riskier behaviours, such as more excessive loss patterns or time spent gambling, than observed among Internet sports bettors and poker players. However, we expect to find moderate and consistent gambling among the majority of the population with a small minority (i.e. 5% or less) exhibiting excessive gambling behaviour.
This article describes the actual Internet casino gambling behaviour of a large cohort of participants during 2 years of a longitudinal study. We established a research cohort and accumulated their subsequent casino gambling transactions at a gambling website. The cumulative information base of these transactions documents each player's gambling behaviour at that site. Although previous investigations examined the self-reported betting activity of Internet gambling on various types of games,34 this is the first study to document the actual gambling behaviour of Internet gamblers playing casino-type games of chance. We present three types of results: (i) an epidemiological description of demographic characteristics of 4222 sequentially subscribed Internet casino gamblers; (ii) an epidemiological description of the casino gambling behaviour of these Internet gamblers and (iii) an epidemiological description of the Internet casino gambling behaviour of an empirically-determined group of heavily involved bettors.
The full research cohort included 48 114 people who opened an account with the Internet betting service provider, bwin Interactive Entertainment, AG (bwin), in February 2005. The majority of subscribers engaged primarily in sports gambling. As expected, relatively few, 8472 (18%), elected to play some casino games, and half of those, 4225 participants, were excluded for playing fewer than 4 days during the study period. The large number of bettors excluded for limited involvement is typical of people curious enough to try the product, but not sufficiently interested to continue casino betting. We also eliminated 10 bettors who played for ‘fun’ (i.e. only played with betting service promotional funds). Finally, we excluded 15 bettors because they had limited exposure to casino play, starting their casino play less than 1 month before the end of the current study period (i.e. between 1 January and 30 January 2007). The longitudinal cohort eligible for the study consisted of 4222 participants.
The available demographic characteristics of the research sample included age, gender, country of residence and preferred language. At enrollment, participants elected to interact with the wagering system in one of 22 languages.
The gambling behaviour measures are based on participants’ monetary deposits to, and withdrawals from, their wagering accounts, as well as daily aggregates of betting activity records. The daily betting aggregates include the number of bets made, total monies wagered and winnings credited to the bettors’ accounts. The daily aggregations provided summary measures of gambling behaviour. We obtained number of bets and total wagered by summing the daily aggregations. We measured the duration of gambling involvement as the number of days from the first eligible bet to the last (i.e. duration). We defined the frequency of involvement as the percent of days within duration that included a bet (i.e. frequency). We obtained the average bets per day by dividing the total number of bets made by the total number of days on which a bet was placed (i.e. bets per day) and the average size of bets by dividing the total monies wagered by the total number of bets (i.e. euros per bet). The net result of gambling (i.e. net loss) is the difference between total wagers and total winnings. The dominant outcome is a net loss and, by subtracting total winnings from total wagers, positive values indicate net losses, the cost of gambling. Converting net losses to a percent of total wagers (i.e. percent lost) provides an index of losses that is independent of the total amount wagered.
We conducted a secondary data analysis of the subscriber database obtained from bwin as described above. We received approval from our Institutional Review Board to conduct this secondary data analyses.
We summarized the participants’ demographics and gambling behaviour using descriptive statistics. Tests for differences between group means included testing the assumption of equal variances and, if necessary, adjusting for unequal variances. We organized the analyses into three sections: (i) cohort characteristics; (ii) cohort gambling behaviour and (iii) the behaviour of heavily involved bettors. For cohort characteristics, we reported gender and country distributions, as well as gambling behaviour differences by gender. For cohort gambling behaviour, we reported gambling involvement by time (i.e. duration and frequency), betting intensity (i.e. number of bets, bets per day, euros per bet), and monetary outcomes (i.e. total wagered, net loss and percent lost). For gambling behaviour, we report medians because of the skewed nature of the gambling data.
The cohort average age was 30 years (SD = 9.0) and most (93%) were male. The players represented 46 countries. The majority indicated residence in Germany (19%), Austria (11%), Greece (11%) and Spain (10%), but substantial proportions of participants were from France (9%), Denmark (8%), Italy (8%), Turkey (8%) and Poland (5%). The remaining 10% of participants were evenly distributed among 37 other countries.
Betting behaviour was similar across genders with the single exception that women placed significantly more bets per day than men (Mwomen = 141, SD = 206 versus Mmen = 114, SD = 191, P < 0.05). Consequently, the data did not justify additional gender-specific analyses.
Internet casino gambling behaviour
The wagering of this cohort on casino games included >206 000 records of daily aggregates that tracked 14.8 million bets, risking €114.7 million and losing a total of €3.5 million. Table 1 summarizes the betting activity for this cohort (N = 4222). The typical number of days of playing casino games is roughly 18; we estimated the number by multiplying the median duration (261 days) by the median percent of days (7%) gambled within the duration (frequency). We estimated the typical cost per day by dividing the median net loss (€117) by the typical number of days of play. Briefly, the central tendencies (medians) describe a cohort that plays casino-type games about once every 2 weeks during a 9-month period and loses about €6.5 at each session. The relationships between the means and the medians, and the size of the standard deviations in relation to the means, indicate that the total distribution is markedly skewed (i.e. extreme betting activity limits the ability of the means to adequately describe the general betting activity of the population majority). We will consider the heavily involved bettors in a later section.
The distributions of the measures violate assumptions of bivariate normality required for product-moment correlations. Consequently, our analysis of the independence among measures used non-parametric rank-order correlation procedures to avoid the undue influence from extreme observations. Table 2 presents the Spearman rank–order correlations between pairs of measures. In large samples, relatively small correlations (in this case, as small as 0.05) are statistically significant. Only one correlation presented in Table 2, the correlation between duration and bets per day, was not statistically significant. Therefore, it is important to consider the size of these correlations as well as their significance.
Correlations among gambling behaviour measures for fixed-odds betting (n = 4222)
No. of bets
Bets per day
Euro per Bet
No. of Bets
Bets per day
Euros per Bet
Duration, interval in days between first and last bet; frequency, percent of days within duration when a bet was placed; net loss, total wagers minus total winnings; Percent lost, net loss divided by total wagered. Non-parametric Spearman correlations all P < 0.001, unless indicated by §
In Table 2, most of the correlations between measures are both significant and large. Participants who wagered larger amounts of money also placed more total bets, more bets per day, wagered more per bet and lost more money overall. Percent lost was negatively correlated with all other measures of betting involvement, indicating that bettors who bet more and more often lost a lower percent of their total wagers than others. Though duration and frequency were highly negatively correlated, indicating that the longer subscribers remained active on the site the lower the percent of days on which they bet, these two measures did not correlate highly with the other measures of gambling behaviour.
Gambling behaviour of heavily involved bettors
We examined subject centile plots to identify empirically whether subgroups within our sample evidenced discontinuously high involvement with casino wagering. Similar to interpreting a screen plot by identifying the ‘elbow’ of that plot, Figure 1 demonstrates for total wagered a discontinuous distribution beginning at the 95th centile. This also was the case for net loss. The total wagered and net loss measures of involvement are highly correlated (Spearman r = 0.70) and two-thirds of the most involved bettors were common to both measures of money at risk. As shown in Table 2, total wagered was correlated more highly with betting activity, both total bets and bets per day, and was considered a better measure of gambling involvement. The temporal measures of duration and frequency were skewed but not markedly discontinuous. We analysed the most top 5% of casino gamblers identified by total wagered (i.e. the top 5%) separately to provide a more complete description of the most heavily involved Internet casino gamblers.
As Table 3 shows, the top 5% of players and their less involved counterparts significantly differed on a number of variables. The single exception was gender. The proportion of females in the top 5% group was lower, 4.2%, compared to the other players, 7.2%; however, this difference was not statistically significant (χ2 = 2.69, P = 0.10). The top 5% players were significantly older by 4 years (t = 6.4, ndf = 233, P < 0.001). The top 5% exhibited significantly increased gambling behaviour compared to other gamblers on all measures of activity and spending. However, the top 5% lost a significantly smaller percent of their total wagers compared to the rest of the cohort (t = 21.0, ndf = 871, P < 0.001).
↵a: All measures significantly different between groups at P < 0.001
Although Internet gambling is often the subject of public health debate and concern, there is little empirical evidence available to inform such debate and address that concern. Stakeholders, however, have speculated about Internet gambling and related public policy in both the popular press and public health circles.6,8–10,26,35,36 Fortunately, empirical data describing population-level Internet gambling behaviour is mounting. Contributing to this growth, this study presents the first ever analysis of real-time betting behaviour of Internet casino gamblers. These findings provide a description of the Internet casino gambling behaviour evidenced by a large cohort of bettors followed prospectively for 2 years. We also identified and reported the characteristics of a distinct group of heavily involved players who comprise five percent of the overall cohort. This information will allow stakeholders to participate in evidence-based public health debate, rather than rely on conventional wisdom and professional speculation.
It is important for public health officials who might be developing Internet gambling-related policy to understand the magnitude of a population's involvement in various types of Internet gambling. We hypothesized that games of chance would not be a popular gambling choice for our longitudinal cohort of sports bettors. During the 2-year study period, 18% of the cohort tried their hand at casino games but half of them did not play on more than 3 days. The finding that only 9% of the cohort played casino-type games to any extent confirms our expectation about the popularity of this gambling option for sports bettors. This finding suggests that, rather than a general interest in Internet gambling, participants are likely to be selective in the types of games that they choose to play.
The service provider that generated the sample of gamblers for the current investigation is most well known for its sports betting services; consequently, it is not entirely clear whether our findings suggest population-level game preferences or indicate a level of specificity only observed among Internet sport gamblers. We noted that females are underrepresented in the longitudinal cohort and this might be the result of gender differences in game preferences. However, gender does not appear to influence actual betting behaviour; neither this study of casino gambling nor the sports gambling study29 observed behavioural differences sufficient to discriminate between genders. Although casino gamblers comprise a small portion of the longitudinal sample, both the full subscriber sample and the subsample of casino gamblers are large (i.e. 4222). The experience of >4000 gamblers observed for as long as 2 years constitutes a significant empirical information base about Internet casino-style games of chance. As we hypothesized, the typical daily cost of casino gambling is modest, but considerably larger than the sports betting costs of this cohort. As we noted in the Results section, the typical daily cost of gambling on casino games was €6.5 per day which is larger than the €1.2 typical daily cost of gambling on fixed-odds sports propositions and the €0.8 typical daily cost for live-action bets.29 However, the cohort of casino bettors played less frequently than the sports bettors. Casino bettors played about twice a month (median frequency = 7%) compared to about seven times a month for fixed-odds bettors (median frequency = 23%) and live-action bettors (median frequency = 27%).29 The observation that casino game bettors incur larger losses at each gambling session compared to sports bettors is consistent with our hypothesis that casino-type games offer an additional risk for players.
The correlation analyses provide important insights about general patterns of Internet gambling behaviour. The high correlations exhibit the consistency of casino betting patterns among these bettors. The correlation between the total number of bets made and the average number of bets per day (Spearman r = 0.87) reflects the day-to-day betting consistency of casino players. The correlation between total monies wagered and net loss (Spearman r = 0.70) is necessarily high because the outcome of casino gambling is a function of chance and the house odds. In our cohort, we also observed a general tendency for rational decision making. The total amount of money wagered correlated negatively with the percent that was lost; wagering decreased as losses increased. Similarly, measures of betting activity and amount per bet also correlated negatively with percent lost. These findings suggest that for this cohort, bad luck was a disincentive for gambling, though more research focused on the temporal nature of these patterns is necessary to confirm this suspicion.
Although many gambling outcomes were uniform (i.e. positively correlated), we also observed a population level split in type of gambling engagement. More specifically, the time involvement measures, duration and frequency, were negatively correlated. This suggests two styles of casino play in our sample: playing on more days during a shorter total play period, and playing less frequently but for a longer period of time. Both play styles had similar outcomes as measured by monies lost. Percent lost correlated negatively with both frequency (Spearman r = −0.18) and duration (Spearman r = −0.07) More frequent play was associated with a smaller percent lost than was a longer duration of play. Although future research is necessary to clarify this issue, our findings suggest that winning reinforces playing on adjacent days more than it reinforces playing over a longer period of time.
Heavily involved players of casino games
Similar to our earlier analysis of Internet sports gamblers,29 the pattern of gambling involvement in this cohort of casino gamblers was discontinuous. A 5% subgroup of the longitudinal cohort (n = 212) was more involved with casino gambling than the rest of the cohort. It is notable that among sport gamblers, discontinuity occurred at 1%;29 hence, a greater proportion of our sample of casino gamblers participated in more extreme gambling behaviour than did sports bettors. It is also worth noting that, on average, the extreme 5% subgroup lost €77 per day compared with the remaining casino players who lost €2.3 per day. All measures reflected this increased gambling involvement for this 5% subgroup. If such groups of heavily involved players indicate noteworthy rates of disorder, behavioural algorithms comprised of temporal, intensity and financial gambling measures might be useful indices for developing website warning systems.
Time involved and money spent
Because financial losses are arguably the most obvious consequence of pathological gambling, common sense suggests that public health attention would feature interventions that concentrate on potential financial and/or material problems and treatment outcomes. However, an equally important consequence of pathological gambling might be how gamblers redistribute their time (e.g. spending less time with family or at work). Defined by financial risk, the extreme 5% subgroup were active casino players during a longer period, they played on more days during the time they were active, and the measure of time spent at casino sessions (i.e. median bets per day) was four times larger than the remainder of the sample. The heavily involved players played frequently, for a long duration, and were recognized by their financial commitment. However, the correlations based on the total sample suggest that some gamblers might experience personal problems unrelated to the amount of money risked. In the full sample, duration and frequency are strongly negatively correlated (Spearman r = −0.61), and both have modest correlations with the Euros per bet (Spearman r = 0.05 and 0.09, respectively). The negative correlation could signal the presence of gamblers who played intensely but for only a few days: an episodic loss of control that could be problematic, but associated with only limited financial losses. The relatively small correlations of duration and frequency with monies wagered could signal the presence of gamblers who spent a long time playing casino games, but did not (or could not) bet more than very small sums. In this case, the time engaged in casino betting, rather than the amount lost, could be the negative outcome of disordered gambling. The time-related findings confirm the suggestion that interventions need to target a range of behaviours and that identification of disordered gambling behaviour needs to move beyond financially related consequences.
Despite the strength of this sample and the research focus on actual gambling behaviour, this study is not without limitations. The observed Internet betting behaviour might not represent a participant's total online gambling behaviour. In addition to playing other types of games on bwin (e.g. sports betting), unlike land-based gambling venues, bettors can access Internet sites easily, play at several venues and move among them readily. The proffered payback rates vary from site to site and it would not be unusual for gamblers to ‘shop’ for the most favourable rates. It also is possible that multiple individuals bet using the same account. The casino games players in this study are a minority (9%) of the longitudinal cohort. The service provider, bwin, is best known as a sports gaming service. It is possible that many gamblers whose primary interest is casino games would select sites that emphasize casino games. The casino players in this sample also bet on sports and might represent bettors with more varied gambling interests than players at sites that emphasize casino games. Although epidemiological information from this and other studies derived from our longitudinal cohort29 advance our understanding of Internet gambling, additional research is necessary to determine how well these findings generalize to other types of Internet gambling. Research has indicated that game preferences at casinos and other land-based gambling venues (e.g. the lottery, bingo, casino games) depend upon the players’ demographic, economic and health-related factors32 as well as cultural and social acceptability.37 Researchers now need to consider whether the observed patterns of games played at land-based gambling venues carry over to Internet gambling.
Games people play
Our data did not provide information about the specific casino games that individuals in our sample played. However, published reports of provider odds might shed light on this issue. The outcomes of casino games are governed by chance with the odds set by the provider. The website indicates that Video Poker and Slots had the lowest returns to the players, overall losses of 6.2 and 5.8%, respectively (https://casino.bwin.com/casino.aspx?view=payoutTable). Casino table games were most favourable with a loss rate of 2.3%, followed by card games with a rate of 2.9%. It is possible that the lower fractional losses for the most heavily involved players (median = 2.5%) are due to a preference for table and card games. Similarly, the higher fractional loss (median = 5.9%) experienced by the large majority of more casual players might be consistent with a preference for slots and video poker. This is an important direction for future research and eventually could suggest directions for targeted public health interventions based on gaming preferences.
The purpose of our research collaboration with bwin is to provide an empirical foundation to guide the development and implementation of strategies that will protect the public health. The rapid expansion of Internet access and services outpaces the acquisition of empirical evidence necessary to develop effective regulations and policies to assure public safety and health. However, an advantage of Internet capabilities is the ability to collect the actual behaviour of a large research sample over a long period of time. This allows research to avoid the nuances of self-report and the prohibitive logistical constraints of repeatedly surveying large samples. This study is a necessary step toward informing the wide range of gambling stakeholders about the behavioural epidemiology of Internet gambling on casino-type games. Research must next begin to identify the population segments at greater or lesser risk for developing Internet gambling-related addiction problems. The determinants for increasing or decreasing the likelihood of developing Internet gambling problems can then serve as a guide for the development of prevention and treatment programs.
Bwin.com, Interactive Entertainment, AG provided primary support for this study. The authors extend special thanks to Christine Thurmond and Ziming Xuan for their support and work on this project. Dr LaBrie had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Conflicts of interest: None declared.
This is the first study to provide evidence about the actual gambling behaviour of a large cohort of Internet gamblers who played casino games during a 2-year period.
The study revealed moderate gambling behaviour at the population level: the median betting behaviour of Internet casino gamblers was to play casino games once every 2 weeks, placing a median of 49 bets of €4 each.
A small percentage of the cohort (i.e. 5%) exhibited behaviours that deviated markedly from the norm: the most involved bettors played every fifth day, placing a median of 188 bets of €25 each.
Two patterns of Internet casino play emerged among the cohort: playing on more days during a shorter total play period, and playing less frequently but for a longer period of time. Little evidence suggested a difference in outcomes across these distinct play styles.
Internet casino gamblers incurred greater daily losses and played less frequently (about twice a month) than Internet sports gamblers (about 7 times per month).
Richard A.LaBrie, Sara A.Kaplan, Debi A.LaPlante, Sarah E.Nelson, Howard J.ShafferEur J Public Health(2008)18 (4):
410-416DOI: http://dx.doi.org/10.1093/eurpub/ckn021First published online: 23 April 2008 (7 pages)