Back tracking: A brief look at using big data in gambling research
I’ve been working in the area of gambling for nearly 30 years and over the past 15 years I have carrying out research into both online gambling and responsible gambling. As I have outlined in previous blogs, one of the new methods I have been using in my published papers is online behavioural tracking. The chance to carry out innovative research in both areas using a new methodology was highly appealing – especially as I have used so many other methods in my gambling research (including online and offline surveys, experiments in laboratories and ecologically valid settings, offline focus groups, online and offline case study interviews, participant and non-participation observation, secondary analysis of survey data, and analysis of various forms of online data such as those found in online forums and online diary blogs).
Over the last decade there has been a big push by gambling regulators for gambling operators to be more socially responsible towards its clientele and this has led to the use of many different responsible gambling (RG) tools and initiatives such as voluntary self-exclusion schemes (where gamblers can ban themselves from gambling), limit setting (where gamblers can choose how much time and/or money they want to lose while gambling), personalized feedback (where gamblers can get personal feedback and advice based on their actual gambling behaviour) and pop-up messages (where gamblers receive a pop-up message during play that informs them how long they have been playing or how much money that have spent during the session).
However, very little is known about whether these RG tools and initiatives actually work, and most of the research that has been published relies on laboratory methods and self-reports – both of which have problems as reliable methods when it comes to evaluating whether RG tools work. Laboratory experiments typically contain very few participants and are carried out in non-ecologically valid settings, and self-reports are prone to many biases (including social desirability and recall biases). Additionally, the sample sizes are also relatively small (although bigger than experiments).
The datasets to analyse player behaviour are huge and can include hundreds of thousands of online gamblers. Given that my first empirical paper on gambling published in the Journal of Gambling Studies in 1990 was a participant observational analysis of eight slot machine gamblers at one British amusement arcade, it is extraordinary to think that decades later I have access to datasets beyond anything I could have imagined back in the 1980s when I began my research career. The data analysis is carried with my research colleague Michael Auer who has a specific expertise in data mining and we use traditional statistical tests to analyse the data. However, the hardest part is always trying to work out which parameters to use in assessing whether the RG tool worked or not. The kind of data we have includes how much time and money that players are spending on the gambling website, and using that data we can assess to what extent the amount of time and money decreases as a result of using limit setting measures, or receiving personalized feedback or a pop-up message.
One of the biggest problems in doing this type of research in the gambling studies field is getting access to the data in the first place and the associated issue of whether academics should be working with the gambling industry in the first place. The bottom line is that we would never have been able to undertake this kind of innovative research with participant sizes of hundreds of thousands of real gamblers without working in co-operation with the gambling industry. (It should also be noted that the gambling companies in question did not fund the research but provided simply provided access to their databases and customers). In fact, I would go as far as to say the research would have been impossible without gambling industry co-operation. Data access provided by the gambling industry has to be one of the key ways forward if the field is to progress.
Unlike other consumptive and potentially addictive behaviours (smoking cigarettes, drinking alcohol, etc.), researchers can study real-time gambling (and other potentially addictive behaviours like video gaming and social networking) in a way that just cannot be done in other chemical and behavioural addictions (e.g., sex, exercise, work, etc.) because of online and/or card-based technologies (such as loyalty cards and player cards). There is no equivalent of this is the tobacco or alcohol industry, and is one of the reasons why researchers in the gambling field are beginning to liaise and/or collaborate with gambling operators. As researchers, we should always strive to improve our theories and models and it appears strange to neglect this purely objective information simply because it involves working together with the gambling industry. This is especially important given the recent research by Dr. Julia Braverman and colleagues published in the journal Psychological Assessment using data from gamblers on the bwin website showing that self-recollected information does not match with objective behavioural tracking data.
The great thing about online behavioural tracking data collected from gamblers is that it is totally objective (as it provides a true record of what every gambler does click-by-click), is collected from real world gambling websites (so is ecologically valid), and has large sample sizes (typically tens of thousands of online gamblers). There of course some disadvantages, the main ones being that the sample is unrepresentative of all online gamblers (as the data only comes from gamblers at one website) and nothing is known about the person’s gambling activity at other websites (research has shown that online gamblers typically gamble at a number of different websites and not just one). Despite these limitations, the analysis of behavioural tracking data (so-called ‘big data’) is a reliable and cutting-edge way to assess and evaluate online gambling behaviour and to assess whether RG tools actually work in real world gambling settings with real online gamblers in real time.
To get access to such data you have to cultivate a trusting relationship with the data providers. It took me years to build up trust with the gambling industry because researchers who study problem gambling are often perceived by the gambling industry to be ‘anti-gambling’ but in my case this wasn’t true. I am ‘pro-responsible gambling’ and gamble myself so it would be hypocritical to be anti-gambling. My main aim in my gambling research is to protect players and minimise harm. Problem gambling will never be totally eliminated but it can be minimised. If gambling companies share the same aim and philosophy of not wanting to make money from problem gamblers but to make money from non-problem gamblers, then I would be prepared to help and collaborate.
You also need to be thick-skinned. If you are analysing any behavioural tracking data provided by the gambling industry, then you need to be prepared for others in the field criticizing you for working in collaboration with the industry. Although none of this research is funded by the industry, the fact that you are collaborating is enough for some people to accuse you of not being independent and/or being in the pockets of the gambling industry. Neither of these are true but it won’t stop the criticism. Nor will it stop me from carrying on researching in this area using datasets provided by the gambling industry.
Dr. Mark Griffiths, Professor of Gambling Studies, International Gaming Research Unit, Nottingham Trent University, Nottingham, UK
Auer, M. & Griffiths, M.D. (2013). Behavioral tracking tools, regulation and corporate social responsibility in online gambling. Gaming Law Review and Economics, 17, 579-583.
Auer, M. & Griffiths, M.D. (2013). Voluntary limit setting and player choice in most intense online gamblers: An empirical study of gambling behaviour. Journal of Gambling Studies, 29, 647-660.
Auer, M. & Griffiths, M.D. (2014). Personalised feedback in the promotion of responsible gambling: A brief overview. Responsible Gambling Review, 1, 27-36.
Auer, M. & Griffiths, M.D. (2014). An empirical investigation of theoretical loss and gambling intensity. Journal of Gambling Studies, 30, 879-887.
Auer, M. & Griffiths, M.D. (2015). Testing normative and self-appraisal feedback in an online slot-machine pop-up message in a real-world setting. Frontiers in Psychology, 6, 339. doi: 10.3389/fpsyg.2015.00339.
Auer, M. & Griffiths, M.D. (2015). Theoretical loss and gambling intensity (revisited): A response to Braverman et al (2013). Journal of Gambling Studies, 31, 921-931.
Auer, M. & Griffiths, M.D. (2015). The use of personalized behavioral feedback for problematic online gamblers: An empirical study. Frontiers in Psychology, 6, 1406. doi: 10.3389/fpsyg.2015.01406.
Auer, M., Littler, A. & Griffiths, M.D. (2015). Legal aspects of responsible gaming pre-commitment and personal feedback initiatives. Gaming Law Review and Economics, 6, 444-456.
Auer, M., Malischnig, D. & Griffiths, M.D. (2014). Is ‘pop-up’ messaging in online slot machine gambling effective? An empirical research note. Journal of Gambling Issues, 29, 1-10.
Auer, M., Schneeberger, A. & Griffiths, M.D. (2012). Theoretical loss and gambling intensity: A simulation study. Gaming Law Review and Economics, 16, 269-273.
Braverman, J., Tom, M., & Shaffer, H. J. (2014). Accuracy of self-reported versus actual online gambling wins and losses. Psychological Assessment, 26, 865-877.
Griffiths, M.D. (1990). Addiction to fruit machines: A preliminary study among males. Journal of Gambling Studies, 6, 113-126.
Griffiths, M.D. & Auer, M. (2011). Approaches to understanding online versus offline gaming impacts. Casino and Gaming International, 7(3), 45-48.
Griffiths, M.D. & Auer, M. (2015). Research funding in gambling studies: Some further observations. International Gambling Studies, 15, 15-19.
Watch this space: Another look at box-set bingeing
Regular readers of my blog will know that I have both a professional and personal interest in ‘box set binging’ – people like myself who sit and watch a whole television series at once either on DVD or on television catch-up services (see my two previous articles on the topic here and here). In my previous blogs on the topic I noted there was a lack of published academic research on the topic. However, a new study on the phenomenon – ‘Just one more episode’: Frequency and theoretical correlates of television binge watching’ – has just been published by Emily Walton-Pattison and her colleagues in the Journal of Health Psychology. The paper argues that binge watching may have detrimental health implications and that binge watching has impulsive aspects. As the authors noted in their paper:
“With the emergence of online streaming television services, watching television has never been so easy and a new behavioural phenomenon has arisen: television binge watching, that is, viewing multiple episodes of the same television show in the same sitting. Watching television is the most widespread leisure-time sedentary activity in adults (Wijndaele et al., 2010), involving little metabolic activity (Hu et al., 2003). In the United Kingdom, over one-third of adults spend at least four hours a day watching television (Stamatakis et al., 2009). Up to 33% of men and 45% of women in the United Kingdom fail to achieve recommended physical activity levels (Craig and Mindell, 2014). As lack of physical activity is the fourth leading mortality risk factor (World Health Organization, 2010), identifying factors that pre- vent achieving health-protective levels of physical activity remains important Furthermore, sedentary behaviour is linked with adverse health outcomes independently of physical activity (Veerman et al., 2012). Time spent watching television is also linked with obesity and reduced sleep time (Vioque et al., 2000). Understanding the factors that lead to watching television at ‘binge’ levels may help to target interventions to reduce sedentary activity and obesity rates and improve sleep hygiene”.
The study involved 86 people who completed an online survey that assessed (among other things) outcome expectations (assessed via six attitudinal items such as ‘Watching more than two episodes of the same TV show in the same sitting over the next 7 days will lead me to be physically healthier’), proximal goals (assessed via one question ‘On how many days do you intend to watch more than two episodes of the same TV show in the same sitting over the next 7 days?’), self-efficacy (assessed via five attitudinal items such as ‘I am confident that I can stop myself from watching more than two episodes of the same TV show if I wanted to’), anticipated regret (assessed via two items – ‘If I watched more than two episodes of the same TV show in the same sitting in the next 7 days, I would feel regret’ and ‘If I watched more than two episodes of the same TV show in the same sitting in the next 7 days I would later wish I had not’), goal conflict (with two items such as ‘How often does it happen that because of watching more than two episodes of the same TV show in the same sitting, you do not invest as much time in other pursuits as you would like to?’), goal facilitation (assessed via three items such as ‘Watching more than two episodes of the same TV show in the same sitting in the next 7 days will help/facilitate my participation in regular physical activity’), and self-reported binge watching over the last week (defined as “watching more than two episodes of the same TV show in one sitting”), as well as noting various demographic details (age, gender, marital status, number of children, and body mass index).
The study found that their participants reported binge watching at least once a week (an average of 1.42 days/week) and that binge watching was predicted most by intention and outcome expectations. Automaticity, anticipated regret, and goal conflict also contributed to binge watching. Based on their results, the authors noted:
“The findings have implications for theory development and intervention…The role of automaticity suggests that interventions aiming to address problematic binge watching (e.g. due to increased sedentary activity) could consider techniques that address automaticity. For example, some online streaming services include in-built interruptions after a number of consecutive episodes have been viewed. There would be opportunities to harness these interruptions. Goal conflict findings indicated that participants who reported more binge watching also reported that binge watching undermined other goal pursuits. Linking such findings to an intervention addressing anticipated regret could provide a useful opportunity…Drawing upon the addiction literature in relation to other types of binge behaviours may further refine potential appetitive and loss of control features that may extend from addictive behaviours with a binge potential, such as eating, sex and drugs, to binge watching”.
Obviously the study relied on self-reports among a small sample of television viewers but given that this is the first-ever academic study of binge watching, it provides a basis for further research to be carried out. As in my own research into gambling where we have begun to use tracking data provided by gambling companies, the authors also note that such objective measures could also be used in the field of researching into television binge watching:
“[Future research] could include using objective measures of binge watching including ecological momentary assessment, ambient sound detection, recording and/or partnering with streaming firms or software-based monitoring. Further insight into binge watching could make a distinction between television show-specific factors, such as genre, length, real-time versus on-demand services, as well as contextual factors (e.g., where binge watching occurred, with whom and when) and assess the association between binge watching and health outcomes including physical activity, eating and sleep hygiene”.
This is one of the first times I can end one of my articles by saying that this is literally a case of “watch this space”!
Dr Mark Griffiths, Professor of Behavioural Addictions, International Gaming Research Unit, Nottingham Trent University, Nottingham, UK
Bates, D. (2015). Watching TV box-set marathons is warning sign you’re lonely and depressed – and will also make you fat. Daily Mail, January 29. Located at: http://www.dailymail.co.uk/health/article-2931572/Love-marathon-TV-session-warning-sign-lonely-depressed.html
Craig, R. & Mindell, J. (2014). Health Survey for England 2013. London: The Health & Social Care Information Centre.
Daily Edge (2014). 11 signs of you’re suffering from a binge-watching problem. Located at: http://www.dailyedge.ie/binge-watching-problem-signs-1391910-Apr2014/
Griffiths, M.D. (1995). Technological addictions. Clinical Psychology Forum, 76, 14-19.
Hu, F.B., Li, T.Y., Colditz, G.A., et al. (2003) Television watching and other sedentary behaviors in rela- tion to risk of obesity and type 2 diabetes mellitus in women. JAMA, 289, 1785–1791.
Kompare, D. (2006). Publishing flow DVD Box Sets and the reconception of television. Television & New Media, 7(4), 335-360.
Spangler, T. (2013). Poll of online TV watchers finds 61% watch 2-3 episodes in one sitting at least every few weeks. Variety, December 13. Located at: http://variety.com/2013/digital/news/netflix-survey-binge-watching-is-not-weird-or-unusual-1200952292/
Stamatakis, E., Hillsdon, M., Mishra, G., et al. (2009) Television viewing and other screen-based entertainment in relation to multiple socioeconomic status indicators and area deprivation: The Scottish Health Survey 2003. Journal of Epidemiology & Community Health, 63, 734–740.
Sussman, S., & Moran, M.B. (2013). Hidden addiction: Television. Journal of Behavioral Addictions, 2(3), 125-132.
Veerman, J.L., Healy, G.N., Cobiac, L.J., et al. (2012) Television viewing time and reduced life expec- tancy: A life table analysis. British Journal of Sports Medicine, 46, 927–930.
Vioque, J., Torres, A. & Quiles, J. (2000) Time spent watching television, sleep duration and obesity in adults living in Valencia, Spain. International Journal of Obesity, 24, 1683–1688.
Walton-Pattison, E., Dombrowski, S.U. & Presseau, J. (2016). ‘Just one more episode’: Frequency and theoretical correlates of television binge watching. Journal of Health Psychology, doi:1359105316643379
Wijndaele, K., Brage, S., Besson, H., et al. (2010) Television viewing time independently predicts all-cause and cardiovascular mortality: The EPIC Norfolk study. International Journal of Epidemiology, 40, 150–159.
Loss leaders: What is the best way to measure ‘gambling intensity’?
The issue of how to measure ‘gambling intensity’ is an important one in the gambling studies field. Gambling intensity is one of those concepts that means different things to different researchers but basically refers to how absorbed gamblers are based on the time and money they spend gambling. Over the last few years, this issue has become much more to the fore as researchers in various jurisdictions have been given access to behavioural tracking data (i.e., actual data showing what online gamblers actually do online such as the games they are playing, the time they spend online, the amount of money that they spend, etc.). This has initiated a whole new line of gambling research that is already providing insights about gambling that we never had before.
Many of these studies have used proxy measures for gambling intensity including variables such ‘bet size’ and ‘number of games played’. Another major problem with these studies is that they have tended to present data by single game type (e.g., only data from online poker players or sports bettors are presented). However, as researchers such as myself have noted, online gamblers typically gamble on a variety of games.
There are various ways to conceptualize gambling intensity. Such ways could include parameters involving the time spent gambling, the number of gambles made, and/or the amount of money won or lost while gambling. In almost all of the studies carried out to date, monetary involvement has tended to be the main proxy used measure for gambling intensity. However, I and my colleague Michael Auer have proposed a different proxy measure for the money risked while gambling. We define gambling intensity as the amount of money that players are putting at risk when playing. This might be considered easy to do (e.g., by using ‘bet size’), but the element of chance is rarely accounted for, especially when a random win occurs. For instance, two gamblers putting the same amount of money at risk might end up with very different wins or losses at the end of similar length gambling sessions because of the chance factor. For this reason, we are now using a measure that is completely independent of random events and takes into account the true amount of money that players are prepared to risk. The interesting aspect of this is that most of the time, gamblers themselves are probably not aware of the amount of money they risked at the end of a playing session.
Our first published paper in this area was a simulation study published last year in the journal Gaming Law Review and Economics. In that paper, we demonstrated that the most robust and stable measure for ‘gambling intensity’ is what we call the ‘theoretical loss’. Our fiest paper on this topic showed that all previous studies using proxy measures for ‘gambling intensity’ had failed to take into account the house advantage. Outcomes in games of chance over the long-term will always be dependent upon the house advantage of each different type of game. Dr. S. Li showed in a 2003 paper published in the Journal of Risk Research that ‘at risk’ decision-making in the short-term is totally different from decision-making over longer periods of time. Decision making over the long-term can be explained by the expected value whereas short-term decision-making does not seem to be based on any expectation rule. However, studies investigating decision-making in situations where people have to make choices assume that players have a real choice in which they can truly influence the outcome and (thus) the expected return. However, this is not the case in pure chance games. Whatever the player chooses to do in pure chance situations, the house advantage will determine the expected return in the long-term.
As we pointed out in our 2012 paper, games with a high house advantage lead to higher player losses and games with a low house advantage lead to lower player losses. Theoretical loss is the same measure that the gaming industry describes as Gross Gaming Revenue (GGR), and is the difference between ‘Total Bet’ and ‘Total Win’. The ‘theoretical loss’ of any given game is represented by the product of the bet size and the house advantage. Over very long periods of time, the theoretical loss corresponds to the GGR with increasing accuracy. The more diverse the gambling behaviour, the more that bet size deviates from the theoretical loss.
By incorporating the theoretical loss, the amount risked can be measured at a very detailed level. For instance, French roulette has a house advantage of 2.7% and keno has a house advantage of 10%. This means that a player who repeatedly bets $100 on roulette will end up with a loss of $2.7, and a player who repeatedly bets $100 on keno will end up with a loss of $10. Therefore, the product of bet size and theoretical loss represents the amount of money that player will lose in the long run. Previous studies that have used bet size (as a proxy measure for gambling intensity) would assign the same gambling of $10 intensity to the two players in the aforementioned example (and which obviously is not the case). The bet size is the one risk parameter that players are most likely to be aware of during gambling. However, it is deceptive as it does not take into account the expected return/loss that is controlled by the gaming operator via their house advantage.
Our simulation study of 300,000 online gamblers showed that bet size explained only 56% of the variance of the theoretical loss, and the number of games played explained 32% of the variance of theoretical loss. This means that when using bet size alone, 44% of the gambling behaviour remains unexplained. When using the number of games played alone, 68% of the variance is left unexplained. As this study was a simulation, we recently replicated our first study using real online gambler behavioural tracking data. There are many advantages and disadvantages with using data collected via behavioural tracking. However, the main advantages are that behavioural tracking data (a) provide a totally objective record of an individual’s gambling behaviour on a particular online gambling website, (b) provide a record of events and can be revisited after the event itself has finished, and (c) usually comprise very large sample sizes.
Our latest study on theoretical loss in the Journal of Gambling Studies comprised 100,000 online gamblers who played casino, lottery or poker games during a one-month period on the Austrian win2day gambling website. All games played by these gamblers were recorded and subsequently analysed. The game types were categorized into eight distinct groups: (i) Lottery – Draw/Instant, (ii) Casino – Card, (iii) Casino – Slot, (iv) Casino – Videopoker, (v) Casino – Table, (vi) Casino Other, (vii) Bingo and (viii) Poker. For each of the game types and each player, the ‘bet size’ and the ‘theoretical loss’ were computed for the recorded time period. In terms of house advantage these game types are very different. In general, lottery games have a relatively high house advantages (typically 50%) whereas slot machines have house advantages in the range of 1 to 5% depending on the gaming platform and the specific game. Poker on the other hand does not have a house advantage as such. In poker, the gaming involvement can be measured via the rake. The rake is a fixed percentage of the stake (bet size) that goes to the casino. The overall theoretical loss is thus comprised of the theoretical loss across all game types plus the poker rake.
Although we found a high correlation between the ‘bet size’ and the overall ‘theoretical loss’ across the eight game types for the 100,000 players, we also found the bet size alone explained only 72% of the variance of the theoretical loss (not as large as we found in our simulation study but that was most likely because we had more games in the simulation study and the games in the simulation study were approximated house advantages whereas the follow-up study used actual house advantages.
This study broadly confirmed the findings from our previous simulation study. The results of our most recent study suggest that future research and particularly those that utilize behavioural tracking approaches should measure their participants’ gambling intensity by incorporating the game-specific theoretical loss instead of using proxy measures such the bet size and/or the amount of money staked. Another implication is that previously published research could be re-analysed using the more robust measure of gambling intensity presented here (i.e., theoretical loss) rather than the proxy measures that were used in the original published studies. This study demonstrates that bet size does not reliably indicate the amount of money that players are willing to risk as it does not take into account the house advantage of each individual game that gamblers engage in. The house advantage represents the percentage held back by the gaming operator and is essential for the amount lost in the long-term and will eventually be equal to the total losses that a player accumulates. In order to further generalize our results, further empirical research utilizing data from other online gaming platforms as well as land-based casino premises needs to be carried out.
Dr Mark Griffiths, Professor of Gambling Studies, International Gaming Research Unit, Nottingham Trent University, Nottingham, UK
Additional input: Michael Auer
Auer, M. & Griffiths, M.D. (2013). An empirical investigation of theoretical loss and gambling intensity. Journal of Gambling Studies, in press.
Auer, M., Schneeberger, A., & Griffiths, M.D. (2012). Theoretical loss and gambling intensity: A simulation study. Gaming Law Review and Economics, 16, 269-273.
Broda, A., LaPlante, D. A., Nelson, S. E., LaBrie, R. A., Bosworth, L. B. & Shaffer, H. J. (2008). Virtual harm reduction efforts for Internet gambling: effects of deposit limits on actual Internet sports gambling behaviour. Harm Reduction Journal, 5, 27.
Colbert, G., Murray, D., Nieschwietz, R. (2009). The use of expected value in pricing judgements. Journal of Risk Research, 12, 199-208.
Griffiths, M.D. & Auer, M. (2011). Online versus offline gambling: Methodological considerations in empirical gambling research. Casino and Gaming International, 7(3), 45-48.
Griffiths, M.D. & Whitty, M.W. (2010). Online behavioural tracking in Internet gambling research: Ethical and methodological issues. International Journal of Internet Research Ethics, 3, 104-117.
LaBrie, R.A., Kaplan, S., LaPlante, D.A., Nelson, S.E., & Shaffer, H.J. (2008). Inside the virtual casino: A prospective longitudinal study of Internet casino gambling. European Journal of Public Health, 18, 410-416
LaPlante, D. A., Schumann, A., LaBrie, R. A., & Shaffer, H. J. (2008). Population trends in Internet sports gambling. Computers in Human Behavior, 24, 2399–2414.
Li, S. (2003). The role of Expected Value illustrated in decision-making under risk: Single-play vs multiple-play. Journal of Risk Research, 6, 113-124.
Wardle, H., Moody, A., Griffiths, M.D., Orford, J. & and Volberg, R. (2011). Defining the online gambler and patterns of behaviour integration: Evidence from the British Gambling Prevalence Survey 2010. International Gambling Studies, 11, 339-356.