Assessing the effectiveness of a responsible gambling behavioural feedback tool for reducing the gambling expenditure of at-risk players

Abstract

The current study assessed the utility of a responsible gambling (RG) tool that provides players with behavioural feedback about their gambling. Data was obtained from 779 people (n = 694 male; n = 85 female) who gambled online with Svenska Spel (the Swedish gambling operator) and who opted to receive behavioural feedback via an RG tool (Playscan). Importantly, data was also obtained from a matched sample of 779 players who did not opt to receive behavioural feedback. Feedback took the form of a colour-coded risk rating (Green = no issues, Yellow = at-risk, Red = problematic), which was determined by a proprietary algorithm. Additionally, gambling expenditure data (amounts deposited and wagered) was gathered for the week in which players enrolled to use the RG tool, the subsequent week and 24 weeks later (this data was also gathered for the matched sample). Results showed that Yellow (i.e. at-risk) players who used the tool significantly reduced the amounts of money deposited and wagered compared to players who did not use the tool – an effect observed the week following enrolment as well as 24 weeks later. Thus, informing at-risk players who have opted to receive feedback about their gambling appears to have a positive impact on subsequent expenditures.

Keywords::

• responsible gambling
• behavioural feedback
• tools

Traditionally, the responsibility to avoid problematic play has been focused on the player. More recently, however, gambling operators have increasingly assumed a duty of care for their patrons (Wohl, Sztainert, & Young, 2013). As a result, many gambling operators have taken steps to create policies and tools that help protect players from the potential harms (e.g. financial, familial, professional) that excessive gambling may cause. Within the Internet gambling environment, responsible gambling (RG) tools can capitalize on behavioural tracking, whereby individual patterns of playing behaviour are monitored for changes that might suggest the development of risky play. Such data can, theoretically, be used to provide players with information about their play, including whether or not the intensity of their play has changed. There are now several behavioural tracking tools that claim to determine when a given player's behaviour has increased in risk and that providing such feedback positively impacts subsequent RG behaviour.

In the current research, we examined the utility of one such RG behavioural tracking tool that tracks players' play on specific Internet (gambling) games and provides them with feedback about how ‘risky’ their playing appears. This was achieved by examining whether the amount of money deposited as well as the amount of money wagered changed after receiving such feedback. In addition, we examined whether the feedback provided to players by the RG tool influenced deposit and wager amounts according to how risky their play was at the onset. Specifically, we hypothesized that the RG feedback would be effective for players who were showing some increase in risky play, but not necessarily among those for whom their play was determined to be highly risky at the onset.

Responsible gambling tools via the Internet

Technology has been a major influence in the field of RG over the last 15 years. Developments in gambling technology mean that many electronic gambling platforms now provide players with, among other RG tools, the opportunity to self-exclude, information about support services, and warning messages about the harms that can result from problematic play (e.g. Auer & Griffiths, 2013; Bernhard, Lucas, & Jang, 2006; Monaghan, 2008, 2009; Monaghan & Blaszczynski, 2007, 2010a, 2010b; Nisbet, 2005; Sharpe, Walker, Coughlan, Enersen, & Blaszczynski, 2005; Stewart & Wohl, 2013; Williams, West, & Simpson, 2007; Wohl, Christie, Matheson, & Anisman, 2010; Wohl, Parush, Kim, & Warren, 2014; Wood & Bernhard, 2010; Wood & Griffiths, 2008; Wood & Griffiths, 2014; Wood & Wood, 2009).

In a recent assessment of extant responsible gambling tools, Wood, Shorter, and Griffiths (2014a, 2014b) identified that there are roughly twice as many RG tools that can be deployed for Internet gambling games compared to games offered in traditional gambling venues (e.g. in casinos, bingo halls, betting shops). As such, they argued that Internet gambling has the potential to be played more responsibly than games offered in traditional ways. In support of this argument, Philander and Mackay (2014) found that past-year online gambling predicted a decrease in problem gambling severity. Moreover, in a study of positive playing habits, Wood and Griffiths (2014) found that a majority of players reported it was easier to stick to their spending limit when buying lottery tickets and virtual scratch cards via the Internet than from a physical retail outlet (e.g. a convenience store). Players also reported that for all other Internet games played it was neither easier nor harder to stick to their limits when playing via the Internet compared to traditional gambling venues (e.g. casinos, racetrack, bingo halls, etc.). The relationship between electronic gambling and problem gambling, however, is not clear-cut and may be dependent, at least in part, on both the structural characteristics of specific games as well as the RG tools that accompany them (Wood & Griffiths, 2014).

Using behavioural tracking data to support responsible gambling on the Internet

It has been argued that there is an increasing emphasis within the gambling industry on players being assisted and encouraged to make well-informed decisions about their gambling behaviour (Bernhard, 2007; Reith, 2008). Within the gambling industry, informed player choice is now a major focus of most responsible gaming policies and strategies (Wood & Bernhard, 2010). The essential requirements for informed player choice have been discussed at length by Blaszczynski, Ladouceur, Nower, and Shaffer (2005), who espouse that (1) the player is personally responsible for his or her level of participation in gambling; (2) players can only make responsible decisions if they are well informed; and (3) science can contribute in determining which information is necessary to promote informed choice in gambling. Central here is the provision of information to players regarding how much time and money has been spent over a particular gambling session or sessions. Such information should encourage players to consciously monitor their playing behaviour and make adjustments when they are required.

Internet-based gambling, by its very nature, means that gambling operators are provided with a plethora of account-based and behavioural information about their players (e.g. amounts deposited, bet and wagered; frequency of play; duration of play; engagement with related RG tools). Such detailed information may have utility in enhancing informed choice amongst players. In a review of the potential for studying account-based gambling, Gainsbury (2011) argued that in order to understand gambling behaviour it is essential to examine these behavioural patterns of play and that there is good evidence that past behaviours can be used to predict future gambling actions and potential problems (Lam & Mizerski, 2009). Similarly, Auer and Griffiths (2014) suggested that providing personalized feedback about actual gambling behaviour, through tracking player behaviour, should have greater utility for informed player choice than generic information about best practices (e.g. adherence to a preset monetary limit).

However, an issue for investigating the utility of providing personal feedback to players as a RG strategy is how problematic gambling can accurately be identified. For example, it has been noted that current problem gambling screens may not be entirely suitable for use in the detection of problematic Internet gambling. Dragicevic, Tsogas, and Kudic (2011) pointed out that problem gambling screens typically classify chasing behaviour as returning the following day to gamble (e.g. Diagnostic and Statistical Manual of Mental Disorders IV (DSM-IV; American Psychiatric Association, 1994), the Canadian Problem Gambling Index (Ferris & Wynne, 2001) and the South Oaks Gambling Screen (Lesieur & Blume, 1987)). However, some Internet gambling players engage in multiple gambling sessions within the same day. This means that chasing behaviour could take place during a much shorter time period. Dragicevic and colleagues (2011) also suggested that chasing on Internet gambling may be more accurately identified by observing increasing variability of bet amounts following losses, rather than observing the bet variability for a specific day or gambling session. As such, it might not be possible to use behavioural tracking data to accurately identify whether a particular player meets DSM-IV criteria for disordered gambling (or whether someone is at risk for developing a gambling disorder).

Nevertheless, we contend that, at the most basic level, behavioural tracking in the gambling domain might be used to identify patterns of play that may correlate with disordered gambling profiles. In support of this contention, LaBrie and Shaffer (2011) examined account data from 679 sports betters, including 215 former players who had closed their accounts and cited problem gambling as the reason. They found that people identified as likely to have a gambling problem typically bet more frequently, placed larger bets and played more intensely shortly after enrolment. In a review of the literature, Delfabbro, King, and Griffiths (2012) concluded that tracking frequency, duration and intensity of gambling, betting patterns and raising funds to gamble (e.g. borrowing money) may predict whether a player has developed disordered patterns of play. Thus, although behavioural tracking data may not function well as a standardized screen for disordered gambling, there is reason to believe that there would be symmetry between risky individual playing patterns as determined by effective behavioural tracking software and disordered gambling as recognized by traditional indices, scales or diagnostic interviews (e.g. the Problem Gambling Severity Index (Ferris & Wynne, 2001)). In as much as behavioural tracking data can be used in lieu of traditional disordered gambling indices, scales or diagnostic interviews, such information becomes intriguing from an RG perspective. This is because behavioural tracking data may allow gambling operators to examine (a) whether their players are gambling responsibly (or) in a risky or problematic manner, (b) whether the introduction of an RG tool influences behaviour, and (c) whether some games are more associated with risky play than others.

There is some limited evidence to support the utility of providing players with personalized feedback as an effective RG strategy. Cunningham, Hodgins, Toneatto, Rai, and Cordingley (2009), for example, randomly assigned problem gamblers to receive (1) full feedback, constituting personalized feedback and normative feedback (i.e. information about how the average player gambles); (2) personalized feedback only; or (3) no feedback. Personalized feedback was compiled after participants completed a questionnaire that was mailed to them at 3-, 6- and 12-month intervals. Personalized data was based upon self-report responses to questions about frequency of play on different games, amounts spent gambling, attitudes about gambling and belief in luck or wining systems, chasing behaviour, guilt, stress, perceived perceptions of significant others (concerning gambling behaviour) and financial problems. The study found that 12 months after baselines were established, players who only received the personalized feedback gambled on significantly fewer days than those who received both personalized feedback and normative feedback and the control group who received no feedback. The study suggests that personalized feedback may be helpful to players with some gambling-related issues. However, the study relied upon self-report data for the personalized feedback, rather than actual observations of gambling. As such, the results were dependent upon participants truthfully and accurately recalling their past behaviours. Consequently, whilst it appears that it may be possible to positively change players' behaviour through personalized feedback, to date there has not been a study that has demonstrated this using the observed behaviour of real players in a naturalistic setting. The question for the present study was: does receiving feedback about a potentially risky change in Internet gambling behaviour, significantly alter subsequent playing behaviour and (if so) in what ways?

Overview of the current study

The current study examined whether a personalized feedback tool (Playscan) that is based on behavioural tracking data promotes responsible gambling. We did so by examining player weekly deposit amounts as well as wager amount data obtained from the Swedish state-operated gambling company Svenska Spel (the company that owns Playscan). Importantly, this company uses an RG tool that tracks behaviour and then informs players whether they are playing problematically (use of the tool, at the time of the study, was voluntary). Information about play is provided to the player using a traffic light metaphor. Players who are defined by the system (via player tracking data) as gambling without a problem are given a green light. Players who are judged to be engaging in some risky play are given a yellow light. When play is defined as becoming problematic, players are given a red light.

To assess the RG utility of the RG tool, we examined whether use of the tool influenced changes in weekly deposit and wager amounts. Moreover, we examined whether the tool's RG utility differed as a function of the player's risk rating (i.e. Green, Yellow or Red) at enrolment. To this end, we also obtained weekly deposit amount and amount wagered data from a matched group of players on the same gambling website who did not enrol to use the RG tool. It is important to note that the behaviour of all players on the website was tracked, but only those enrolled to use the RG tool received behavioural feedback. Use of a matched control group allowed for assessment of whether receiving personalized feedback influenced play over and above normal variations in behaviour. We hypothesized that personalized feedback via the RG tool's colour scheme would result in a down regulation of subsequent deposit and wager amounts for players who were told they were in one of the two risky play categories (i.e. Yellow or Red). Moreover, this reduction in deposit and wager amounts should be significantly greater than any change in behaviour observed among the matched control group. To this end, we examined possible behaviour change (i.e. deposit and wager amounts) between the week in which players enrolled to receive behavioural feedback and the following week, as well as 24 weeks later.1

Method

Participants and procedure

Approximately 1.5 million players were registered to play Internet games with the Swedish state-operated gaming company Svenska Spel at the time we were given access to their player data. Only residents of Sweden are allowed to register to play Internet games with Svenska Spel. Of the players registered, 65,000 voluntarily registered to receive behavioural feedback via the online tool. Data-mining staff at Playscan undertook the process of sampling from all Svenska Spel players in accordance with the authors' specified selection criteria. Specifically, the authors asked the data miners to match a person who enrolled in Playscan to receive behavioural (BF) with a person who had not enrolled in Playscan and thus received no behavioural feedback (NBF) on age, sex, colour (i.e. risk) category at time of the BF player's enrolment, types of games played, the average amount deposited during the 10 weeks prior to the week of enrolment for the BF player, and the average amount wagered during the 10 week prior to the week of enrolment for the BF player. All matches were included in the final sample. This matching process yielded 779 matched pairs for a total sample of 1558 Internet players (male = 1388; female = 170). Importantly, in addition to data relevant to the matching process, raw data was received that pertained to each player's wagering and deposit amounts for the period of time specified by the authors as well as the risk rating for each specified time period. The authors performed all data analysis (without direction from Svenska Spel).

The BF tool is based on a proprietary algorithm that calculates a risk score based on the intensity of play over a 10-week span. The risk score is then sorted into one of three colour categories (Green, Yellow, Red) that correspond to the intensity of a player's gambling behaviour in relation to his or her previously observed playing behaviours. If the risk score suggests the player is engaged in low intensity or recreational play a green light is displayed. If the risk score suggests a player is engaged in moderately intense or risky play a yellow light is displayed. A red light is displayed if the risk score suggests the player is engaged in very intense or risky play. If there is no gaming data for 10 weeks of play, players are assigned a Gray (no data) status. No players in our sample had a Gray status. Where a player played more than one game type, the riskiest category was recorded; this is because the BF tool assesses individual games rather than cumulatively across several games.

The percentage of participants in each risk category at the start of the study was as follows: Green 80%, Yellow 13% and Red 7% (matched in both BF and NBF groups). Importantly, Week 1 represented the first occasion that a BF participant was informed of his or her colour rating.

Games that participants played (equally amongst both groups) were: Bingo 7%; Lottery 57%; Sport betting 54%; Poker 15%. Obtaining data from those who did not enrol to use the RG tool was possible because all player behaviour is monitored by the tool regardless of whether the player chooses to enrol (and thus receive behavioural feedback). Importantly, all data was de-identified before it was received from the gambling company. As Gainsbury (2011) noted, player data provided to independent academic researchers would not be considered to be in violation of player's privacy provided that this data is de-identified and researchers have no way in which to identify or contact players. Additionally, players provided consent to use their de-identified data as part of the terms and conditions when signing up to use the gambling companies' Internet gambling website.

Dependent variables

Amount deposited

Data was gathered on the amount of money BF group members deposited into their account on the week of their enrolment, the week following enrolment and 24 weeks after enrolment. These two time points were chosen to assess whether knowledge of initial play intensity (via colour category notification) influenced subsequent amounts deposited. This data was also gathered for each BF group's matched NBF group member.

Amount wagered

Wagers for a given player are tallied weekly. We obtained wagering data for players in the BF group on the week of their enrolment, the week following enrolment and 24 weeks after enrolment. Akin to deposit amount, these two time points were chosen to assess whether knowledge of play intensity (via colour category notification) influenced subsequent wagering. Data was also obtained for each BF group's matched NBF group member.

Results

Data analytics

Prior to reporting the results of the current study, a note is warranted about the reported p-values. Due to the large sample size, a given difference between groups may be statistically significant without resulting in practical significance (i.e. statistical significance does not imply a meaningful difference between groups). Although we report the p-value in the results section, to properly interpret findings, we focus on the effect size. In this report, we use the partial eta squared, η², as the effect size measurement. An eta-squared of .01 is indicative of a small effect, an eta-square of .06 indicative of a medium effect, and an eta-square of .14 is indicative of a large effect (see Cohen, 1988).

Change in deposit amount: week of enrolment to following week

To assess whether receiving behavioural feedback (i.e. whether a player was classified Green, Yellow or Red and informed of this categorization) was associated with changes in deposit amount in the week following a BF player's enrolment, we first created a deposit amount change score. Subtracting the deposit amount at week of enrolment by the deposit amount one week later created the change in deposit index. Thus, positive numbers indicate a net increase in deposit amount. We repeated this for each matched member of the NBF group. We then examined the change score for outliers (i.e. change scores that were greater than 3SDs above or below the mean). All outliers (N = 27) were removed for subsequent analyses.

A 2(Group: BF vs. NBF) × 3 (Feedback category at week of enrolment: Green vs. Yellow vs. Red) between-participants ANOVA was then conducted with change in deposit from week of enrolment to the following week as the dependent variable. Results revealed no significant main effect of group, F(1, 1526) = .64, p = .45, η_p < .001, or feedback category, F(2, 1526) = .73, p = .48, η_p = .001. A small but not meaningful interaction between group and category qualified the lack of main effects, F(21,526) = 3.39, p < .03, η_p = .004.

Due to our a priori hypotheses, we decided to break down the interaction by behavioural feedback category even though the effect size was very small. Results revealed no significant differences in deposit amount between the Red BF (M = 3844.23, SD = 50,046.83) and Red NBF players (M = − 5109.43, SD = 43,479.43), F(1, 103) = .96, p = .33, η_p = .009. Similarly, there were no significant differences between Yellow BF (M = − 8264.89, SD = 4245.63) and Yellow NBF players (M = 323.16, SD = 41,839.25), F(1, 187) = 1.84, p = .18, η_p = .01. There was, however, a meaningful difference between Green members of the BF group and Green members of the NBF group in terms of change in deposit amount, F(1, 1236) = 19.28, p < .001, η_p = .02. Specifically, there was a greater net reduction in deposit amount among Green BF group members (M = − 7518.37, SD = 29,538.13) compared to Green NBF group members (M = − 1138.36, SD = 20,915.19).

Lastly, we ran a one-way ANOVA to assess possible between category changes after receiving behavioural feedback (i.e. people in the BG group). Results revealed no significant differences according to behavioural feedback category, F(1, 758) = 2.85, p = .06, η_p = .007.

Change in deposit amount: week of enrolment to Week 24

To assess any possible longer-term (i.e. 24-week) impacts of receiving behavioural feedback on amounts deposited, we first created a deposit amount change index by subtracting the deposit amount at week of enrolment from the deposit amount 24 weeks later. Thus, positive numbers indicate a net increase in deposit amount. We repeated this for each matched member of the NBF group. We then examined the change score for outliers (i.e. change scores that were greater than 3SDs above or below the mean). All outliers (N = 25) were removed for subsequent analyses.

A 2(Group: BF vs. NBF) × 3 (Feedback category at week of enrolment: Green vs. Yellow vs. Red) between-participants ANOVA was then conducted with change in deposit from week of enrolment to Week 24 as the dependent variable. Results revealed significant but small difference between groups, F(1, 1527) = 8.24, p = .004,  = .005. BF players reduced their deposit amount from week of enrolment to Week 24 (M = − 13,828.74, SD = 35,326.88) more than NBF players (M = − 3094.55, SD = 31,044.72). As predicted, there was also a significant but small difference between feedback categories, F(2, 1527) = 7.43, p = .001,  = .01. Yellow players showed the greatest change in their deposit amount (M = − 16,576.96, SD = 50,006.01). This was significantly different from the reduction observed among Green players (M = − 7010.08, SD = 28,021.50), p = .001, but not significantly different from Red players (M = − 10,438.23, SD = 51,934.08), p = .28. There was no significant difference between Red and Green category players, p = .57.

Importantly, a significant and meaningful interaction between group and category qualified these results, F(2, 1533) = 10.70, p < .001,  = .02 (see Figure 1). Thus, we broke the interaction down by behavioural feedback category. Results revealed no significant difference between BF Red (M = − 26,948.21, SD = 19,3425.44) and NBF Red players (M = − 12,553.57, SD = 13,4840.71), F(1, 100) = 1.34, p = .25,  = .01. There was, however, a significant difference between groups among Yellow and Green players, F(1, 189) = 12.96, p < .001  = .06 and F(1, 1238) = 44.21, p < .001,  = .03, respectively. Yellow BF players reduced their deposit amount from Week 0 to Week 24 to a greater extent (M = − 29,412.77, SD = 50,445.65) than Yellow NBF players (M = − 4138.14, SD = 46,542.52). Similarly, Green BF players (M = − 12,219.39, SD = 29,919.77) reduced their deposit amount to a greater extent between Week 0 and Week 24 than Green NBF players (M = − 1817.55, SD = 24,954.74).

Figure 1 Deposit change (Swedish Kroner): week of BF enrolment to 24 weeks later.

Lastly, we ran a one-way ANOVA to assess possible between category changes following behavioural feedback (i.e. people in the BF group). Results revealed significant differences according to behavioural feedback category, F(1, 759) = 11.92, p < .001, η_p = .03. Post-hoc tests revealed that BF yellow players reduced their deposit to a greater extend (M = − 29,412.77, SD = 50,445.65) than both BF Red (M = − 4263.27, SD = 52,223.81) and BF Green players (M = − 12,219.39, SD = 29,919.77), ps < .001. There was no significant difference between BF Red and BF Green players, p = .27.

Wager amount analyses: week of enrolment to following week

To further assess whether behavioural feedback category knowledge impacts gambling behaviour, we examined the possibility that enrolment in the behavioural feedback system influenced the amount of money wagered in the week following enrolment. Subtracting the wager amount at week of enrolment from the wager amount one week later created the change in wager index. Thus, positive numbers indicate a net increase in wager amount. We repeated this for each matched member of the NBF group. We then examined the change score for outliers (i.e. change scores that were greater than 3SDs above or below the mean). All outliers (N = 32) were removed for subsequent analyses.

A 2(Group: BF vs. NBF) × 3 (Feedback category at week of enrolment: Green vs. Yellow vs. Red) between-participants ANOVA was then conducted with change in amount wagered from week of enrolment to subsequent week wagering as the dependent variable. Results revealed a significant difference between groups, F(1, 1520) = 10.30, p = .001, ηp < .007. Players in the BF group reduced their wager amounts (M = − 20,513.08, SD = 20,4836.67) to a greater extent than players in the NBF group (M = − 7423.92, SD = 10,9249.26). As predicted, there was also a significant but small change according to feedback category, F(2, 1520) = 10.79, p < .001, η_p < .01. Yellow players showed the greatest change in their deposit amount (M = − 57,336.97, SD = 26,9037.19). This was significantly different from the reduction observed among Green players (M = − 4904.64, SD = 12,5056.50), p < .001, but not significantly different from Red players (M = − 43,497.64, SD = 27,351.12), p = .77. Like Yellow players, Red players reduced their wagering more than Green players, p = .05.

A significant but meaningless interaction between group and category qualified the main results, F(2, 1520) = 4.60, p = .01, η_p = .006. In fact, when we broke down the interaction by feedback category no significant results were obtained. Specifically, no significant differences were observed between BF Red (M = − 75,694.74, SD = 37,4286.59) and NBF Red players (M = − 13,685.52, SD = 11,5699.70), F(1, 102) = 1.34, p = .25, η_p = .01, BF Yellow (M = − 75,694.74, SD = 37,4286.59) and NBF Yellow players (M = − 13,685.52, SD = 11,5699.70), F(1, 102) = 1.34, p = .25, η_p = .01, or Green BF (M = − 5614.47, SD = 15,8818.91) and Green NBF players (M = − 4202.79, SD = 78,657.72), F(1, 1235) = .04, p = .84, η_p < .001.

Lastly, we ran a one-way ANOVA to assess possible between category differences after receiving behavioural feedback (i.e. people in the BF group). Results revealed a significant change according to behavioural feedback category, F(1, 751) = 9.12, p < .001, η_p = .02. Post-hoc tests revealed that BF yellow players reduced their wagering to a greater extent (M = − 92,463.21, SD = 31,0669.84) than BF Green players (M = − 5614.47, SD = 15,8818.91), p < .001, but not BF Red players (M = − 75,694.74, SD = 37,4286.59), p = .89. BF Red players also reduced their wagering to a greater extent than Green players, p = .89.

Wager amount analyses: week of enrolment to Week 24

To assess the possibility that behavioural feedback category knowledge impacts gambling behaviour over the long term, we examined whether enrolment in the behavioural feedback system was associated with the amount of money wagered 24 weeks following enrolment. Subtracting the wager amount at week of enrolment from the wager amount 24 weeks later created the change in wager index. Thus, positive numbers indicate a net increase in wager amount. We repeated this for each matched member of the NBF group. We then examined the change score for outliers (i.e. change scores that were greater than 3SDs above or below the mean). All outliers (N = 30) were removed for subsequent analyses.

A 2(Group: BF vs. NBF) × 3 (Feedback category at week of enrolment: Green vs. Yellow vs. Red) between-participants ANOVA was then conducted with change in amount wagered from week of enrolment to subsequent week wagering as the dependent variable. Results revealed a significant difference between groups, F(1, 1522) = 14.03, p < .001, η_p = .01. Players in the BF group reduced their wager amounts (M = − 43,638.11, SD = 27,0199.07) to a greater extent than players in the NBF group (M = − 8005.92, SD = 19,4340.61). There was also a significant but small change according to feedback category, F(2, 1522) = 6.36, p = .002, η_p = .009. Yellow players showed the greatest change in their wager amount (M = − 79,340.52, SD = 41,9846.79). This was significantly different from the reduction observed among both Red (M = − 4202.59, SD = 41,1716.55) and Green players (M = − 19,306.49, SD = 16,5818.29), ps < .02. Red and Green players did not differ in their wager change amount, p = .80.

A significant but small interaction between group and category qualified the main results, F(2, 1522) = 7.24, p = .001, η_p = .01 (see Figure 2). Thus, we broke the interaction down by behavioural feedback category. Results revealed no significant difference between BF Red (M = − 28,535.46, SD = 40,7203.49) and NBF Red players (M = 18,753.22, SD = 41,8505.09), F(1, 101) = .34, p = .56,  = .003. There was also no significant difference between BF Green (M = − 27,826.29, SD = 20,7982.30) and NBF Green players (M = − 10,868.87, SD = 10,8572.63), F(1, 1236) = 3.24, p = .07,  = .003. There was, however, a significant change by group among Yellow players, F(1, 185) = 6.64, p = .01  = .04. Yellow BF players reduced their wager amount from Week 0 to Week 24 to a greater extent (M = − 16,0251.41, SD = 46,2086.68) than Yellow NBF players (M = − 4268.58, SD = 36,2958.88).

Figure 2 Wager change (Swedish Kroner): week of BF enrolment to 24 weeks later.

Lastly, we ran a one-way ANOVA to assess between category changes after receiving behavioural feedback (i.e. people in the BF group). Results revealed significant differences according to behavioural feedback category, F(1, 753) = 9.74, p < .001, η_p = .03. Post-hoc tests revealed that BF yellow players reduced their wagering to a greater extend (M = − 16,0251.41, SD = 46,2086.68) than both Red (M = − 28,535.76, SD = 40,7203.49) and Green players (M = − 27,826.29, SD = 20,7982.30), ps < .02. There was no significant difference between BF Red and Green players in change in wagering, p = .99.

Discussion

The purpose of the current study was to assess whether presenting behavioural feedback to players on the extent of their play has responsible gambling utility. Specifically, we sought to examine whether an RG tool that informs players about their play (especially if that play is classified as risky) could have behavioural consequences: that is, a reduction in the amount a player deposits as well as wagers in subsequent weeks. We found that players who received behavioural feedback (the BF group) showed a significantly greater reduction in amounts deposited one week after receiving behavioural feedback compared to those who did not receive behavioural feedback (the NBF group). Whilst Red members of the BF group (i.e. players who were defined as showing signs of problematic play) and Yellow (i.e. players who show signs of risky play) did not significantly reduce their deposits for this period, Green (i.e. players who show signs of problem-free play) members of the BF group did show a significant reduction in their deposits for this period. Furthermore, these deposit reductions were noticeable over time. Green and Yellow BF players both showed a significant reduction in amounts deposited from week of BF enrolment to Week 24 compared to NBF players and this was most distinct for Yellow BF players. Again, there was no difference in the amounts deposited between red BF and NBF players.

Interestingly, Green BF players were found to decrease the amounts deposited and may have been adjusting their playing behaviour independently of the BF feedback. One speculative explanation for this is that their interest in using a BF tool suggests that they were more sensitized to monitoring their gambling behaviour than the Green NBF group. However, if that were the case it is not possible to determine whether actually using the BF tool impacted their deposit limits. That is, did it sensitize them to monitor their deposits more so than the NBF Greens, or were they in fact already doing this? Either way, we might argue that the tool is meeting a need for these players, as evidenced by their opting in; however, the mechanism for the efficacy of the tool, at this level, is less clear.

Red players in both the BF and NBF group also reduced their wagering to a similar extent between the week of BF enrolment and the subsequent week and the same effect was observed when comparing enrolment with Week 24. This may reflect an attempt by both BF and NBF Red players to consciously change their behaviour. Conceivably, Red players might be more aware of the potentially risky nature of their playing behaviour (than other categories) as their playing patterns are, by definition, more extreme. Yellow members of the BF group significantly decreased their wagering, compared to both Green and Red players. This was observable both at the week following enrolment and again 24 weeks later. This again suggests that the BF tool could potentially have the most impact on at-risk players.

Overall, use of the BF tool appeared to reflect a desired outcome according to the current conceptualization of responsible gambling (Blaszczynski, Ladouceur, & Shaffer, 2004; Wood & Griffiths, 2014). That is, the main focus should be upon assisting at-risk players (i.e. Yellow) to regain control over their playing behaviour, whereas Red (potentially problematic) players may already be at the point where RG efforts will be less helpful than more serious interventions such as treatment referral and self-exclusion.

Limitations of the study

This study is one of the first to show, using actual observed Internet gambling data, that an online BF tool is linked to what may be regarded as a reduction in risky gambling behaviours (i.e. decreased amounts deposited and wagered by players defined as at risk). However, the question remains as to the extent to which players who use a BF tool are, in any case, more likely to self-monitor and/or be concerned about taking action if their gambling behaviour becomes risky. That is, are players who agree to use an RG tool inherently more responsible than those who choose not to use such a tool? Further research might examine whether there are any differences in attitudes to gambling and self-control, between those who decide to use a BF tool and those who choose not to.

It is important to realize that this study was not an experiment conducted in a laboratory, but rather an attempt to objectively observe the possible impact of behavioural feedback on real players in a real-life setting. The matched samples in the study were not randomly assigned to the BF and NBF groups. It is impractical (and perhaps unethical) for a gaming company randomly to submit some of their customers to a condition that they did not choose (i.e. receiving behavioural feedback). Furthermore, such random assignment of participants into two groups, in a real gambling setting, could conceivably alter the behaviour of players who did not want such feedback. Consequently, direct causal relationships are not possible from this study; however, strong associations were observed. Replication of the study, preferably with a sample of players who are already mandated to receive BF (as gaming company policy rather than for the sake of a research study) would be helpful to observe the extent to which these findings are robust and reliable over time. Similarly, it would be useful to replicate the study with a different population of players to judge the extent to which the findings are applicable to other countries.

Another possible limitation of the study is that there was no way to verify independently the extent to which the risk classifications (i.e. Green, Yellow Red) conform to established categories of problem and at-risk gambling. The assumption is that those BF players who reduced their deposit amounts and wagering, upon being informed of a negative change in their risk rating (e.g. Green to Yellow), were responding to a recognition that their gambling behaviour was becoming a negative issue for them. However, what if such players were receiving false positives (i.e. their behaviour was not becoming risky) and were adjusting their gambling behaviour needlessly? This seems unlikely, given that Griffiths, Wood and Parke (2009) previously found that almost two-thirds (63%) of users in a previous evaluation study (of the same tool) reported that the feedback that they received helped them to be better informed about their gambling behaviour. Presumably, if their classification did not reflect their playing behaviour, at least somewhat, then they would be less than satisfied in this respect.

It is also worth noting that current problem gambling screens are not well suited for identifying disordered gambling in behavioural data sets. Most existing screens focus on self-reports of total gambling behaviour, and usually also examine erroneous cognitions and the consequences of excessive gambling. In contrast, the BF tool examined in the current research identified and communicated actual observed behaviour that was tracked on specific game play. We argue that this is a major benefit of behavioural tracking data, as it does not rely on players accurately and truthfully recalling past play. Self-report screens must rely on subjective and largely retrospective responses from people who may feel confused and possibly stigmatized by their gambling issues. By contrast, some of the advantages of behavioural measurements of risk are that they can be more objective and do not depend upon accurate recall and authentic insights. Also, as noted earlier, it has been questioned whether or not existing problem gambling screens are suitable for identifying chasing behaviour on Internet gambling (e.g. Dragicevic et al., 2011). Further research may nevertheless wish to examine the degree to which behaviourally identified risk categories map on to our existing understanding of how to detect problem and at-risk gambling or whether they are measuring something else entirely. However, in terms of informing players about changes in their gambling behaviour, communicating standardized risk categories may be less important than allowing players the opportunity to judge for themselves whether their behaviour is becoming an issue of concern. Empowering individuals to make well-informed choices is, after all, the cornerstone of current thinking about how to facilitate responsible gambling. In this respect, the notion of risk may be codified as relative to the individuals' previous patterns of play rather than specifically mapping on to the much broader criteria of existing problem gambling screens, which tend to assume that gambling problems can be uniformly identified in all populations.

Given the high degree with which BF tool users had previously rated the utility of the tool for keeping them informed about the potential riskiness of their playing behaviour (Griffiths, Wood, & Parke, 2009), it would appear to be meeting a need, at least amongst voluntary users. A challenge for such BF tools will be the extent to which all players can be motivated to engage with such a tool. Again, it would be interesting to see how a BF tool impacts the playing behaviour of all players where use of the tool is a mandatory requirement. That is, to what extent would players who are less motivated to use a BF tool benefit from receiving feedback about their playing behaviour? Another aspect of the BF tools that was not examined in this study was the type of messages and actions that might be presented when a player moves from Green to Yellow, or Yellow to Red. We currently know very little about the optimal message content for motivating RG, although self-appraisal messages have been shown to be helpful (Monagahan & Blaszczynski 2010b). Nor do we empirically know the best type of action to take (e.g. referral to support services, cease all marketing, offer player the option to self-exclude, etc.). Further research could also examine the extent to which players move in and out of categories over time. However, whilst this may have some merit in describing the fluidity of such behaviours, it remains to be seen if it would provide a better understanding of how personalized feedback impacts at-risk players' overall gambling expenditures.

The high number of male participants compared to female participants in this study may somewhat limit the generalizability of the study, and future studies may wish to examine in more detail whether demographics are a factor relating to the efficacy of behavioural feedback. Also, the results obtained are indicative of people playing on one particular website and one game rather than cumulative across several website and games. The impact of some players possibly gambling on multiple websites is difficult to ascertain. We might speculate that feedback from one website might sensitize players' awareness of their overall gambling behaviours. However, where players play several different games on a website that uses the tool, those games that utilize BF can provide snapshots of their overall gambling habits. Different game types have been shown to have different risk potentials for problematic play (Griffiths et al., 2009; Parke & Griffiths, 2007) and it may be possible that some players experience issues with some games but not others. In the present study it was not possible to analyse the impact of the tool across different game types due to the nature of the group matching process to which the researchers were blinded. However, by observing the risk scores (yielded by a BF tool) of players on different types of games, gaming operators and researchers should be better able to determine the extent to which a specific game is associated with a high level of problematic play. In such a scenario, the game might be modified to reduce the risk potential, and/or additional RG features might be included with the game (see Wood et al., 2014a, 2014b, for a further discussion of this strategy).

Conclusion

Overall, results of the current research suggest that the use of an Internet-based BF tool, that informs players that their gambling behaviour is becoming risky, is associated with a reduction in the subsequent amounts of money deposited and wagered by those players. The challenge for gambling operators might be to design, test and implement optimal configurations for such tools as part of their RG development process (see Wohl et al., 2014, for a discussion). The extent to which such actions might help reduce overall levels of problem gambling amongst BF tool users remains to be seen. Further advances in this area will require ongoing cooperation between researchers, BF tool developers and gaming operators to work together toward a common goal of problem gambling harm reduction.

Conflicts of interest

Funding sources: The study was funded by Playscan AB Sweden, the owners and operators of the Playscan behavioural tracking tool.

Competing interests: A conflict of interest might be perceived considering that the funders of the project have a vested interest in the results of the study. However, to guard against such a conflict, the following steps were taken by the research team. Beyond approving the original research question, Playscan had no say in the analysis of the data undertaken, the interpretation of the results or the content of the final report.

Constraints on publishin

Prior agreement between the research team and the funders guaranteed no restrictions upon the publication of the findings.

Additional information

Notes on contributors

Richard T.A. Wood

Richard T. A. Wood is a psychologist who specializes in the study of gambling behaviour. He is the president of GamRes Limited, an international research and consultancy company that develops, evaluates and implements responsible gambling initiatives. He investigates the individual causes of problem gambling as well as the structural characteristics of games that can influence gambling behaviour. He has developed cutting-edge responsible gambling tools such as GAM-GaRD, that informs the development of socially responsible game design, as well as www.GamTalk.org that provides free online support for people with gambling-related issues.

Michael J.A. Wohl

Michael J. A. Wohl is a professor of psychology at Carleton University. The majority of his work has focused on factors that predict problematic gambling behaviour, while his current focus includes factors that facilitate responsible gambling and barriers to treatment-seeking among disordered gamblers. This research is conducted, in part, in the Carleton University Gambling Laboratory. He has published over 80 peer-reviewed papers and is the receipt of Carleton's Research Achievement Award.

Notes

^1. Twenty-four weeks was chosen as this was deemed to be a suitable period to examine the longer-term influences of the tool, whilst short enough to complete the study within the agreed time-line of one year.