Gambling Commission Machines data – more evidence, but any more clarity?

The machine data provided by the Gambling Commission earlier this month provides the most detailed data yet published on machine session behaviour by machine category and venue. However, there is no ‘customer view’, nor is it clear whether certain characteristics are clear markers of harm or not. There is far more that the data does not tell us than it does. Various parties have drawn inferences from the dataset (with concomitant policy implications) – not always in accord with one another. In our view, there are a number of useful observations to be made when the dataset is considered within the context of broader research. This is our take on what can (and cannot) be interpreted and how significant the insights may be.

Earlier this month, the Gambling Commission released data on machine play for substantially all LBOs and a significantly smaller proportion of AGCs and bingo clubs. In theory, there is therefore more evidence upon which to base “evidence-based policy making”. Given that we are in the middle of the most wide-ranging and politically charged review of machine gambling regulation since 2005, this is in one sense timely. However, the data is presented in aggregated form and, understandably, without interpretation from the Commission (which has stated that it has not formed a view on its meaning yet).

The data has already been deployed in battle, for once more thoroughly by the ABB than critics. However, given stakeholder predispositions, the extent of the data and what is at stake (quite literally), some of the conclusions drawn have been inevitably somewhat selective.

As with all stakeholders and commentators, we do not pretend to believe that this data holds the answer to the big policy questions, especially since there is no view of the player (these are session statistics) and in any event harm is a subjective, environmental issue only partially (and sometimes misleadingly) realised in expenditure data. However, we believe the data does provide insights into four key areas:

  • The impact of the “£50 journey” on B2 roulette play
  • The impact of stake size on duration and spend
  • The frequency of “heavy loss” sessions by product and venue
  • The duration of play by product and venue
We therefore consider each of these issues in turn, through three lenses:
  • What the data tells us (along with any issues with drawing conclusions from the data)
  • What this is likely to mean from a behavioural stand-point
  • How this might impact policy

The impact of the £50 journey
The impact of the £50 journey is significant because it provides insight into the effectiveness of the first major regulatory intervention on B2 machines (in 2014); one which many bookmaker stakeholders resisted at the time but now hold up as regulatory best practice. If it is working, then the Triennial Review should logically be relatively benign for B2 but could potentially impose tougher regulatory requirements on other machine types, borrowing from the ABB Code (eg, session time and spend warnings etc).

But what does “working” look like? DCMS articulated the policy objective in January 2016 as:

“The policy objective of these regulations is to assist people who use sub-category B2 gaming machines (commonly known as fixed odds betting terminals or FOBTs) to stay in control of their gambling behaviour by requiring that those accessing higher stakes (over £50) load cash via staff interaction or use account based play.” 

The data shows that the £50 journey has caused an almost exact trade-off between time and money: ie, customers have shifted play patterns substantially away from staking above £50 to place substantially more bets at lower staking levels, particularly above £20 and especially £40-50 (this was already known).

At the same time, however, mean session duration has increased and – importantly – net expenditures appear to have gone up. The likelihood of a machine player in a betting shop incurring losses greater than £200 per session has decreased slightly; but the proportion of players incurring losses greater than £100 per session has increased from 6.2% to 7.2%. While this increase may indicate a change in player behaviour, the increase cannot be treated as statistically significant due to the size of sample; and overall spend has not materially decreased.

The overall outcome is therefore relatively clear: players are taking longer to spend the same sums per session (sessions lasting greater than 30m up 26%), apparently caused by reduced stake per spin. For this to be a positive regulatory outcome, then DCMS needs to be relaxed about overall expenditure on machines and an increase in duration. It is important to point out that overall session expenditure is not in itself a measure of harm (net expenditure per player is not available in the data and even LBO customers have very different income bases), nor is spending longer on a machine necessarily a cause for concern. However, it is hard to state from the data provided, that the £50 journey has been a ‘success’ relative to DCMS policy objectives – players have modified their behaviour, but arguably only to ‘get around’ the restrictions to achieve the same spend. This supports the conclusion from the recent Forrest & McHale study into the effectiveness of the £50 regulations, which concluded: “If the effect of the Intervention was to lead to a pool of players (likely to include many problem gamblers) changing behaviour but still spending the same amount of money as before and extending the time spent at the machines, it seems unlikely that the Intervention achieved its ultimate goal of harm mitigation.”

Of potentially separate significance is that £50 Journey displacement was only visible above £10 stakes and only significant above £20. This would appear to reinforce an anecdotal view that roulette play requires staking above £20 per spin to be a viable product, with a proportionate revenue impact on stake reduction up to £50. In other words, this data would suggest a £50 maximum stake would be a relatively minor disruption to the bookmakers (since the majority of customers are already effectively observing it), whereas anything below a £20 stake would have a very material detrimental impact on roulette revenue to the extent that it is unlikely to be an appealing product (especially) for higher value customers.

The impact stake size on duration and spend
Perhaps unsurprisingly, there is a linear relationship between session length and average stake on B2 machines (ex-slots; i.e. roulette), which is most pronounced at either end of the trading day (early/late).

There are likely to be several underlying behavioural reasons for this:
  • ‘heavier’ players are more likely to spend longer on sessions
  • Higher staking patters are likely to yield more wins, extending session times
  • Winning customers are potentially recycling at higher stakes (increasing average staking, extending duration)

Based upon customer reaction to the £50 journey, however, it would be dangerous to assume that a cut in stakes would reduce session time, even if that were considered a desirable outcome: customers are likely to modify behaviour to achieve a given session spend, assuming the desired game-play were available (ie, a roulette product which can still appeal to higher value customers).

The frequency of “heavy loss” sessions by product and venue
Defining and comparing “heavy loss” sessions is probably the most contentious and risky exercise possible from the available data. Probably the most important factor to consider is that research has demonstrated that the incidence of problem gambling is significantly higher among the homeless (10x) than the overall population (Sharman; Cambridge, 2014): a “heavy loss” for this cohort could be considered almost any loss. Equally, session losses are not the same as customer losses: the number of sessions and the proportion of winning sessions are key to this and not visible in the data. However, session losses are the best proxy the data provides us for the absolute and relative intensiveness of game play, and therefore there is some (caveated) merit in considering it. In order to draw a line somewhere, we have considered “heavy losses” to be over £100 per session and then analysed this by venue and cohort.

It can be seen that LBOs have a materially greater incidence of “heavy losses” than other venues, especially in spend brackets above £500. However, there may be an explanation for a larger number of lower spending sessions in bingo where machine play tends to be disrupted by bingo play.

Loss-making session data can also be used to examine revenue generated by loss cohorts. We have used the mean average loss per cohort and multiplied this by the number of sessions to get to a ‘gross loss’ figure for loss making sessions across key products and venues.

LBO B2 (ex-slots)
  • 0.1% generated from individual session losses over £5,000 (too small to show in the chart)
  • 23% revenue generated from sessions over £500
  • 66% revenue generated from sessions over £100
  • 81% revenue generated from sessions over £50
  • No revenue generated from individual session losses over £5,000
  • 3% revenue generated from sessions over £500
  • 35% revenue generated from sessions over £100
  • 58% revenue generated from sessions over £50
Bingo B3
  • No revenue generated from individual session losses over £5,000
  • 4% revenue generated from sessions over £500
  • 43% revenue generated from sessions over £100
  • 66% revenue generated from sessions over £50
  • No revenue generated from individual session losses over £5,000
  • 7% revenue generated from sessions over £500
  • 55% revenue generated from sessions over £100
  • 75% revenue generated from sessions over £50

It can be seen from the lossmaking session cohorts that “heavy loss” sessions (NB, not the same as net spend per customer) were much more significant to B2 content than any other machine content. B3 characteristics were relatively similar in LBOs and Bingo, with AGCs (which almost entirely rely on machine income for revenue), somewhat more dependent.

It is important to reiterate that these do not equate to player losses, but they are likely to be a reasonable proxy. It is equally important to reiterate that “heavy losses” are not in themselves a marker of harm. However, the fact that B2 content stands out so much is likely to attract the attention of policy makers, especially since it is stake (or more specifically the ability effectively to play roulette) which is the key differentiator between B2 content and other compared content. Further, clear evidence of session losses of this magnitude (seen to some extent in all machines) could also attract attention from an AML / KYC perspective, especially given the ‘typical’ demographic of machine users.

The duration of play by product and venue
B2 content does not stand out in terms of session duration. Indeed, B2 is slightly skewed toward shorter sessions. Much has been made about the difference in sessions over 2 hours, where the incidence is 67% greater in AGCs. However, sessions over 2 hours are only 0.2% of B2 play post the £50 journey (and this has increased by 34% since the £50 journey) and only 0.36% of AGC play.

Duration is therefore a double-edged sword for LBOs: if the ABB’s stance that longer duration in ‘competitor’ venues is an issue, then the £50 journey could be seen to be a material exacerbating factor to a potential harm marker. If, on the other hand, duration is seen as more nuanced, then there are probably few points to score from the data on either side of the debate.

Can any conclusions safely be drawn?
There are too many issues with the data (in isolation) to draw firm conclusions, especially the lack of a ‘player view’ (though this lack may in itself be worthy of policymaker consideration). However, based upon the analysis above, we believe we can draw six key points with a relatively high degree of confidence:

  • Customer responses to the £50 journey have increased duration and not reduced spend; this cannot necessarily be considered a “bad” thing, but it is not an easily identifiable “good” thing either (especially if ABB comments are taken to infer that longer duration is “bad”)
  • £50 journey displacement (and the distribution of “heavy losses” by stake size) suggests that £20 is the lower limit of viable roulette play for higher value customers, and even this is likely to be significantly disruptive if it were adopted as a (hard or soft) maximum stake, though the product still works well at a (soft) £50 limit
  • High staking games and play is clearly correlated to longer duration, though whether this means greater harm is being caused is open to interpretation, while lowering stakes could potentially increase duration (as above); however, duration has been seen as a potential marker of harm
  • Sessions involving “heavy losses” are prevalent in all forms of Category B content and venue, though B2 content has a materially higher gross revenue footprint from heavy lossmaking sessions; given the broadly similar demographics (other than gender) of casual landbased gambling, both of these points this may become a key area of focus for policymakers on a number of levels (potential harm markers, KYC / AML)
  • While “extreme behaviour” (losses, duration) appears small as a percentage of sessions, it is not only the likely to be the area of focus, but also much more important in revenue terms than perhaps hitherto appreciated by some stakeholders
  • The very incompleteness of the picture (especially from a customer perspective) is likely to highlight the need for greater visibility (inevitably more research, potentially player tracking etc)