To the authors’ knowledge, the present study is the first to ever examine the application of the DSM-5 criteria of gambling disorder (APA, 2013) to actual gambling behaviour using online gambling transaction data. Initially (in the Introduction), each of the nine DSM-5 criteria were examined to see how these could be operationalized using account-based tracking data. This was done by consulting with experts from the online gambling industry, regulatory bodies, reformed problem gamblers, and other gambling researchers., as well as utilizing examples from the gambling studies literature After the DSM-5 criteria had been operationalized, the gambling behavioural indicators were analysed utilizing a sample of gamblers who had registered with the gambling operator Unibet from September 1 to December 31, 2018. For each gambler, the gambling activity in their first three months was analysed and operationalized in relation to the DSM-5 criteria for gambling disorder.
Initially, all the operationalized gambling disorder criteria were observed in the sample and the data were presented descriptively. In the second stage, the data were z-normalised to allow comparability and a two-step cluster analysis was performed. Four clusters emerged: non-problem gamblers (646 participants; 65.78%), financially vulnerable gamblers (three participants; 0.31%), emotionally vulnerable gamblers (nine participants; 0.92%), and at-risk gamblers (324 participants; 32.99%). In the first cluster (non-problem gamblers), all the gamblers had a negative z-score for all the proposed gambling disorder criteria. This showed that the majority of the population had values that were smaller than the means when compared to the other clusters.
The second cluster (financially vulnerable gamblers) comprised only three gamblers. In this cluster, it was evident that the number of daily deposits and the increase in daily deposits over time (i.e., chasing behaviour) were at much higher than compared to the mean values of the other clusters. The number of hours spent gambling online, the number of different active days gambled on, and the number of registered credit cards on the account were also higher than the mean values of the other clusters, but not as high as the number of daily deposits and the increase in daily deposits over time. On the other criteria, they were all close to the other cluster mean values, and in some measures (removal of RG tools, cancelled withdrawals, abusive communication, third-party calls, and frequency of bonus begging) were lower than the mean.
The third cluster (emotionally vulnerable gamblers) comprised nine gamblers. In this cluster, the most distinctive behavioural attributes were abusive communication with customer services staff and frequency of bonus begging. Six of the other criteria were negative z-scores, showing that the values were lower than the mean for the other clusters, whereas five of the criteria showed positive z-scores.
The final cluster (at-risk gamblers) comprised one-third of the gamblers. This cluster had no distinctive criteria, and all the values ranged from + 1 to − 1, and four of the criteria (number of daily deposits, abusive communication with customer services, chasing losses, and frequency of bonus begging) had negative z-scores, therefore the values were much lower to the total sample’s means. Frequency of bonus begging, abusive communication with customer services, number of daily deposits, and the increase in daily deposits over time were all lower than the mean of the other clusters. The other criteria had a positive z-score, with the number of active gambling days having the highest value of 0.937, showing that the z-score was almost one standard deviation greater than the mean of the total sample.
The benefits in the approach taken in the present study is that the whole population was observed and not just gamblers that chose voluntary self-exclusion (VSE) which has been the case in previous studies (e.g., Braverman & Shaffer, 2012; Haefeli et al., 2011; Percy et al., 2016). Griffiths and Auer (2016) highlighted that there are limitations in this latter approach because gamblers may not be using self-exclusion for problem gambling reasons. A recent study by Catania and Griffiths (2021) analysed 7732 gamblers who had used VSE. They reported almost one-fifth of the gamblers used the VSE option even though the players had less than 24 h of activity on their account. Moreover, gamblers who use VSE are treated by most researchers as a homogenous group despite the several differences present.
The majority of the sample in the present study were either in the non-problem gambling cluster (65.78%) or the at-risk gambling cluster (32.99%) comprising 98.77% the participants. In both groups, there were no distinctive criteria that were much higher than the means of the total sample which would be expected because the two clusters comprised a high percentage of the total population studied. This may show that these gamblers were playing within their financial means and not problematically. The remaining two clusters that did display elevated values on DSM-5 criteria for gambling disorder comprised 1.23% of the total sample. In the UK, the most recent British Gambling Prevalence Study reported that 0.9% of the population were problem gamblers using the DSM-IV criteria (Wardle et al., 2011). Therefore, the two clusters which comprise 1.23% of the total sample in the present study may reflect the individuals who are problem gamblers given the similarities in prevalence.
The financially vulnerable gambler cluster showed higher than average means in the number of daily deposits, increase in daily deposits over time (i.e., chasing behaviour), number of hours spent gambling, number of different active days spent gambling, and number of registered credit cards on the account. This group appeared to show much greater levels of preoccupying gambling behaviour based on the amount of time and money spent (Griffiths, 2012). Overall engagement and increase in time spent gambling are key predictors of gambling harm (Gainsbury et al., 2019). A high amount of deposited money has also been associated with gamblers that self-excluded for problem gambling reasons (Ukhov et al., 2021). The gamblers that were in this cluster also showed more indicators of chasing behaviour compared to the other clusters, and this may result in relying on others to provide money to relieve desperate financial situations that were caused by gambling (bailout). Gamblers may use bailout money by depositing it on their online gambling account, in order to chase their losses. In the present study, the number of registered credit cards on the gambling account may be an indicator for bailout, and was also higher compared to the mean of the total sample.
The psychologically vulnerable gambler cluster included two behaviours which were higher than the means of the total sample. These were abusive communication with customer services staff and frequency of bonus begging. These two behaviours may be signs of withdrawal and bailout. Gamblers may use aggressive communication in online gambling chatrooms (Griffiths, 2012), and gamblers may use abusive communication due to psychological and/or emotional strain that occurs due to problem gambling (Haefeli et al., 2011). This strain may be due to the losses being out of control for the gambler, or financial funds running out and therefore not being able to continue gambling. Bonus begging may be used in order for the gambler to get ‘free money’ on their account to be able to chase their losses and use this ‘free money’ as means to gamble when the financial means to do so are running low. The findings presented here may help gambling operators in minimising harm caused by gambling, because online behavioural tracking identify potentially problematic gambling behaviour (Haeusler, 2016). Additionally, using objective data collected by tracking technology, may overcome the weaknesses that are present with self-report data such as lying or the social desirability effect (Griffiths, 2009).
Although the present study may help in the discussion of operationalizing online behavioural tracking data and potential gambling disorder indicators, it does have a number of limitations. The dataset is from only one gambling operator, and this may be limited view since most gamblers use more than one gambling operator, and therefore generalisation to all online gamblers may be difficult to conclude (Auer & Griffiths, 2019; Auer et al., 2020). Some of the indicators developed are arguably quite novel (e.g., frequency of bonus begging) and therefore further research should be performed in relation to their reliability as indicators of potential gambling harm indicators. Another limitation could be that some of the operational definitions of each DSM-5 criterion were arguably narrow. For instance, in the present study chasing was simply defined as an increase in deposits over time. Further research should look at longitudinal aspects to see how these potential problem gambling indicators develop over time. Another approach could be to use self-report data in combination with these potential problem gambling indicators to better understand their prevalence with gambling-related harm.
Despite its limitations, the present study creates a foundation of potential new problem gambling indicators that may be used in conjunction with the collecting of online behavioural tracking data. Furthermore, through analysing a whole gambling population and not limiting it to voluntary self-exclusion as a proxy measure for problem gambling, can further help gambling operators to prevent disordered gambling.