At the Responsible Gambling Council’s Discovery 2011 conference we presented our research and solutions that help lotteries and operators to develop sustainable relationships with gamblers by helping them to make more informed decisions. Part of our session was focused on sharing our latest research findings on the analysis of high-risk gambling behaviours.
Braverman and Shaffer (2010), from Harvard’s Division on Addictions, published How do gamblers start gambling: identifying behavioural markers for high-risk internet gambling last year. The paper is an important research asset as it was the first to analyse actual online gambling behaviour during a gambler’s first month of play to predict gambling-related problems. Their study of live action sports bettors identified a small sub-group of gamblers (2.8%) from the total research cohort (n = 530) who demonstrated high levels of gambling variability and involvement. These gamblers were found to be at higher risk than other gamblers of reporting gambling related problems.
The study was of particular interest to us because it was the first to use actual online gambling data to analyse the first month of play, which allows for the possibility of intervention before gamblers start causing harm to themselves. We wanted to see whether their results for sports betting would be consistent with other gamblers so we recreated the study using two new datasets, casino (n = 546) and poker (n = 575), using the same methodology as Harvard i.e. analysing the first month of play following registration, using the k-means clustering method and the same behavioural markers. Our results showed some similarities to Harvard’s, in that we identified in both casino and poker a small-sub group of gamblers who showed markedly different gambling behaviours compared with the others gamblers during their first month of gambling activity, which in our case was highly variable gambling patterns e.g. see cluster 3 in our casino results.
The study was of particular interest to us because it was the first to use actual online gambling data to analyse the first month of play, which allows for the possibility of intervention before gamblers start causing harm to themselves. We wanted to see whether their results for sports betting would be consistent with other gamblers so we recreated the study using two new datasets, casino (n = 546) and poker (n = 575), using the same methodology as Harvard i.e. analysing the first month of play following registration, using the k-means clustering method and the same behavioural markers. Our results showed some similarities to Harvard’s, in that we identified in both casino and poker a small-sub group of gamblers who showed markedly different gambling behaviours compared with the others gamblers during their first month of gambling activity, which in our case was highly variable gambling patterns e.g. see cluster 3 in our casino results.
We also observed other similarities with Harvard’s results e.g. the majority of gamblers in the research cohorts demonstrated moderate betting patterns. There were also some differences e.g. our casino research identified a sub-group of gamblers who demonstrated high levels of gambling Intensity during their first month compared to the other gamblers. This could be attributed to the nature of the casino games, such as slots and roulette, in that they are more continuous compared with live action sports betting and poker, which could allow for more intensive betting behaviour.
Further opportunities exist to build on Harvard’s and our research, including extending the number of risk factors and also leveraging alternative statistical methodologies. Whilst clustering is a useful machine learning technique for dividing data into meaningful groups, it has some limitations. For example it has trouble clustering data with large outliers, such as skewed non-normally distributed populations such as these datasets. It is also produces more meaningful and natural clusters with the application of greater numbers of variables and sub-clusters e.g. Experian, the credit rating agency, has used the k-means clustering technique to cluster populations into 45 types and 13 groups using 350 measures (Cameron et al, 2005. A new methodology for segmenting consumers for financial services. Journal of Financial Services Marketing, Vol 10, 3, 260-271).
These findings provide further evidence of different gambling patterns and behaviours relative to high risk behavioural markers amongst gamblers. We plan to publish the full findings later in the year. However, when new technologies emerge, such as using behavioural analytics to help gamblers to better self-regulate their gambling behaviours, their usefulness is not always obvious and their uptake is rarely dependent on how well the technology works. Rather, success and uptake depends on whether the ideas behind the technology spread and diffuse, which is why it is important that the industry has the opportunity to debate these ideas in detail. At Discovery 2011 there was significant interest and debate across many sessions in how new technologies can help to better protect vulnerable gamblers. Thanks must go to the Responsible Gambling Council for organising a great platform to enable lotteries, operators, software providers, problem gambling prevention and treatment providers and academics to debate these issues in a very open and collaborative manner.
