Gambling machines research hijacked by the industry
Initial suspicions about the independence of this research programme
The first series of reports of the programme of research on machine gambling commissioned by the Responsible Gambling Trust (RGT) are out. The results were presented to an audience largely consisting of industry representatives at a conference held in London on 10th December. This is important research not least because those of us who have been arguing for something significant to be done about Fixed Odds Betting Terminals (FOBTs) in betting shops were told that no policy decisions would be taken until this research had reported. We were always suspicious about this, suspecting that it was a way of avoiding taking any action. We were suspicious because, as we have repeatedly pointed out, the current British system of funding and commissioning research does not guarantee its independence from the industry. RGT is ‘an industry led body’ and was always unlikely to support research which might threaten the massive profits which the industry now makes from FOBTs. It always seemed likely that the machines research programme would follow the industry lead in focusing on problem play and problem players rather than on problem products. This was confirmed early on in the reports of the research where it is repeatedly stated that the main purpose of the programme was ‘to identify problem gamblers and harmful patterns of play’. The large bulk of the reports, and of discussion at the 10th December conference, has focused on trying to predict problem individuals and sessions of play. Suspicions were intensified because the research has been conducted by a consortium including the much respected NatCen and a company called Featurespace which has a history of contracts with the gambling industry which is represented on its board.
A survey of loyalty cardholders: nearly half were found to be problem or at-risk gamblers and most held negative attitudes towards the product
One of the most interesting reports in the series gives us an initial analysis of a survey, using a number of questions similar to those used in the 2010 British Gambling Prevalence Survey (BGPS), of those holding loyalty cards with Ladbrokes, William Hill or Paddy Power. The response rate was estimated at only 17 to 19% and those who agreed to take part in the survey tended to be more engaged gamblers, but the procedure for weighting the data was able to compensate for that to some degree. The most striking feature of the results to me was the extremely high rate of problem gambling (Problem Gambling Severity Index [PGSI] scores of 8 or more) – a massive 23% – with a further 24% scoring as ‘moderate risk’ (scores of 3 to 7). Rates of problem gambling were particularly high amongst those who were unemployed (38%) and were also high amongst those who gambled on a larger number of forms of gambling, those in the lowest of five income groups, those in England who lived in the most deprived areas, those who were economically inactive because of long-term sickness, and members of BME groups.
Two attitude statements, identical to those used in the BGPS but in this case relating specifically to machine gambling, were also included in the survey. Asked whether it ‘is a harmless form of entertainment’, 54% disagreed (24% strongly) and 29% agreed (only 5% strongly). Asked whether it ‘should be discouraged’, 41% agreed (19% strongly) and 37% disagreed (7% strongly). You really might expect loyalty cardholders to be more positive than that. The majority it seems are not very loyal when it comes to commenting on the product for which they hold a loyalty card!
Industry betting transaction data: large minorities of players lose large amounts of money in short periods of time
Other reports looked at betting transaction data provided by the five big bookmaking companies. The size of the database is huge – more than 6.7 billion bets placed in over 32,000 gambling machines in over 8,000 betting shops! In England, 40% of all these bets were placed in betting shops in the 20% most deprived areas – not surprisingly, since this is where 38% of all the betting shops are located (there was a similar concentration of bets in more deprived areas in Scotland). The average bet size was just over £5 (nearly £6 in the 20% most densely populated areas and nearly £7 in London, and nearly £10 after 10 p.m. at night).
One of the most interesting things is how stake size and expenditure of money and time varied by whether a betting session only involved betting on B2 machines (the FOBTs), only on B3s (faster, low stake, jackpot, slot-type machines), or involved both types of machine. It needs to be borne in mind, however, that it is not always obvious from the data when one ‘session’ finishes and another starts. Hence, some arbitrary decisions had to be made in order to break up the transactional data into ‘proxy sessions’. The method for doing this was complicated and there is a suspicion that it may have resulted in breaking the data up into unrealistically short sessions – the median session length was just under 4 minutes which will seem unrealistic to some (the mean was just over 11 mins). Nevertheless, the results are interesting. Not surprisingly, the mean stake per bet during B2-only sessions was £14, much higher than for other types of session. At the 90th percentile it was £38 and at the 95th percentile £65, showing that a large minority of stakes in such sessions are many times higher than the stakes that are possible on B3 machines. Interestingly, though, when it comes to expenditure in sessions of different types, B2-only sessions were no more costly than B3-only sessions, in terms of either money (a mean loss of just over £6 for both types of session) or time (a mean of just under 10 minutes for both types), with the highest means being for sessions that involved both types of machine (a mean loss of just over £14 and a mean session length of a bit over 20 mins). An indication of the large amounts of money that people are losing in quite short periods of time is given by the figures for, say, the 95th percentile for monetary loss which corresponds to a loss of £105, and the 95th percentile for length of session which corresponds to a bit over half an hour.
Linking transactional and loyalty cardholder survey data: prediction of problem gambling from transactional data is poor (and the report of this part of the programme displays bias)
About 4,000 of the loyalty cardholders who agreed to take part in the survey also agreed that their results could be linked to the betting transactional data. A particular aim was to determine what parameters of machine play best predicted whether someone had a score on the PGSI indicating a gambling problem. The analysis started by testing a ‘baseline’ model based on the ABB (Association of British Bookmakers) social responsibility code introduced a few months ago. The best prediction was provided by an expenditure cut-off of £30 (the £250 limit mentioned in the code was far too high and time expenditure added little to the prediction). But that cut-off was not very successful since the best it could do was to identify 60% of the problem gamblers at the expense of 40% of non-problem gamblers being false positives. Compared to that baseline model, it was possible to improve the prediction of problem gambling by using the best combination of a large number of factors from the transactional data set. These factors included, ‘number of playing days’, ‘maximum weekly winnings’, and ‘average daily loss’. The report says that stake size was notable by its absence but stakes did appear in the list of predictors in the form of ‘average daily player total stake’ and ‘number of stake levels’.
In the discussion of these results the authors say that they have shown that they can predict problem gambling and that there is therefore ‘a bright future for behavioural analytics’ in the area of social responsibility and gambling. This really seems to be over-optimistic and to smack of special pleading. Of course no one expects perfect prediction but even the best was hardly impressive. For example, 50% of problem gamblers could be identified but only at a cost of a false positive rate of 25% amongst non-problem gamblers. Since the latter are much the larger group, this represents an unacceptably high false-positive rate. In other words at even a modest level of sensitivity, specificity is unacceptably low. Further analysis did suggest that many of the false positives might be at-risk gamblers.
Unlike the report on the survey of loyalty cardholders (the first author for which was Heather Wardle of NatCen), that report (first authored by Dave Excell of Featurespace) was noticeably less careful. As well as making a point of noting the absence of stake size in their best predictive model, they say in the discussion that, because no one variable alone was predictive (how could it have been!), ‘The focus should shift away from regulating particular parameters, such as stake size, but take a balanced rounded approach which considers the player, the product and the environment’, a clear reference to the debate about high stake FOBT machines which have been so controversial and which were the stimulus for setting up this programme of research in the first place. This report also repeatedly refers to the skewedness of the data (as if this was a surprise!), ‘a small number of extreme values’, the majority of players exhibiting ‘minimal values’ (as an example, ‘only one in 10 will play 40 times a month’!), and most variables following an exponential curve so ‘large values… are very rare’. That is a coded way of saying: Look how moderate almost all machine play is and how rare is immoderate or irresponsible gambling!
There is much confusion and mistrust about ‘return to player’ (RTP) messages
Amongst a number of other projects which contributed to the overall research programme was a small-scale qualitative study exploring players’ understanding and opinions about RTP information. Although participants were split between those who felt that such information was useful and those who did not, a main finding, perhaps not unexpectedly, was that messages about RTP were mostly found to be unclear and/or inaccessible. There was too much technical language, the mathematical concepts involved were not easy for everyone to understand, and the English-only language was an additional problem for some.
A laboratory experimental study suggests that staking impairs decision-making
This experiment was conducted with 32 regular but non-problem gamblers. The authors fully acknowledge the limitations of an artificial experiment like this, but there were some indicative findings. Decision-making, which is one aspect of ‘executive functioning’, was impaired, although this happened at both stake levels used in the experiment (£20 and £2) when compared with the control condition in which no money was staked. Unexpectedly this occurred both when winning as well as when losing. A conclusion drawn by the first author in his talk at the meeting was that structural characteristics do affect decision-making and that the product therefore does matter.
The British gambling research agenda has been hijacked by the gambling industry
I have made the comment before, that this programme of research, although there is much about it which is very interesting, is a good example of how the gambling research agenda has been hijacked by the gambling industry and those who have been taken in by their overall view of things. One obvious indication of this is the makeup of the organisations that have collaborated to sponsor, oversee and carry out the research. In an attempt to provide some independence from the industry, a Machines Research Oversight Panel was set up. Alex Blaszczynski, a well-known Australian gambling researcher whose views on gambling policy are known to be conservative, industry-supporting ones, chaired the Panel. Representing the Panel at the conference Paul Delfabbro stated that, despite industry support for the machines research programme, it did seem that the work had been independently carried out. This is really to miss the point. The research was probably carried out, and mostly reported, without direct influence by the industry, and that is particularly likely to be true of those parts of the programme led by NatCen. The point is that the research required the collaboration of the industry who provided access to much of the data, and the very nature of the research and the questions being asked are part of a whole framework of ideas about ‘responsible gambling’ which is industry-friendly and not too industry-challenging.
One then has to ask about members of the collaborating research team. NatCen has an excellent record of carrying out independent research and in my opinion is itself above reproach. This can not necessarily be said, however, about the other research partners, particularly Featurespace which has a history of working with the gambling industry and has contracts with the industry. As already noted, that part of the report led by Featurespace was much less cautious in drawing conclusions and Featurespace has been accused of drawing unwarranted, industry-favourable, conclusions in a press release and tweets prior to the conference. The front cover of the printed programme for the conference day bears the Featurespace logo alongside logos for igt (‘The industry’s leading manufacturer of gaming machines’), Rank (owners of Grosvenor Casinos), and Joelson Wilson (a law firm that ‘advises remote gambling operators and suppliers, acting for spread-betting companies, online casinos and major betting websites’).
The likely conclusions that the industry and their supporters will draw from the programme of research so far came out in the final panel discussion at the conference. Matthew Hill of the Gambling Commission summed up for me the whole problem with the way the research programme has been set up by saying that the takeout message was that ‘we should be looking at people not the product’. That seems to be strange coming from the body that is charged with regulating the industry! Richard Glynn, Chief Executive of Ladbrokes, was clearer still in concluding that, ‘it’s not the product’. The keyword for him, to summarise the research findings so far, was ‘complex’ – very convenient of course for an industry making huge profits out of a dangerous activity. Alex Blaszczynski made essentially the same point at the end of his written introduction from the Oversight Panel: ‘… at this stage, it would be inadvisable to rush policies on the basis of these foundational studies… more would be achieved by a strategic approach compared to fragmented, disjointed and potentially costly policies that fail to achieve their objective’ – in other words, don’t do anything about FOBTs now. The final word went to Richard Glynn who, in response to my suggestion that the research agenda had been hijacked by the industry, referred to ‘provocateurs’, amongst whom I presume he counted me as one!