University of Calgary
UofC Navigation

Relationship Between Measures of Gambling Intensity and Risk of Harm

On this page you will find:

1. Creation of Low Risk Limits for Gambling 2. Related Work by Other Research Groups 3. Expert Consultation on Low Risk Limits

 
Creation of Low Risk Limits for Gambling

It is tricky to identify where on the continuum of gambling intensity an individual crosses the threshold from safe to unsafe gambling. One of the first challenges was determining the optimal definition of harm in the context of gambling. Unlike substance abuse, excessive gambling does not have readily identifiable health consequences. We reasoned that individuals who report two or more negative consequences from gambling (across all the potential areas of functioning that could be impacted by gambling - social, family, psychological, medical, legal and financial) were experiencing harm. Possible harms due to gambling include: health problems (e.g., stress or anxiety), financial problems, borrowing money to gamble, having gambling habits criticized by others, feeling guilty about gambling, betting more than one could afford to lose, and feeling that one has a gambling problem.
We next used receiver operating characteristics (ROC) curves to determine the monetary and time limits to best differentiate recreational gamblers from 'at risk' gamblers. The ROC curve is a statistical tool that allows for the identification of an optimal cut point of one measure that predicts the presence or absence of a well-defined target outcome.
ROC curves originated during WWII, where they were used for radar signals. It is useful to think of a radar example to understand how ROC curves were used in our projects:
A radar operator had to listen to radar sounds and distinguish a signal (e.g., torpedo) from background noise (e.g., fish). During a radar operator's shift he reports 8 things: three torpedoes, and five fish. The radar operator correctly identifies two torpedoes (torpedo=torpedo; true positive) and misses one by identifying it as a fish (torpedo=fish; false negative). Sensitivity is defined as the proportion of true positives of all signals. In this case it is the number of torpedoes correctly identified (2) / total number of torpedoes (3), so sensitivity = 0.66. But this is only half of the story. On four occasions the radar operator incorrectly identifies a fish as a torpedo (fish=fish; false positive). On only one occasion does he correctly identify a fish as a fish (fish=fish; true negative). Specificity is defined as the proportion of true negatives of all noise. In this case it is the number of fish correctly identified (1) / total number of fish (5), so specificity = 0.20.
Sensitivity and specificity are complimentary ways of describing the accuracy of prediction. If the radar operator was hyper-vigilant for torpedo sounds then he would identify sounds as torpedoes more often and we would see an increase in both true positives and false positives. Thus sensitivity would increase, but specificity would decrease.
In our project we wanted to determine a suitable threshold of gambling intensity that would maximize our ability to differentiate between recreational gamblers and 'at risk' gamblers. Since there were no reasons to favor maximizing either sensitivity or specificity over the other, we decided that the gambling intensity level where both sensitivity and specificity were collectively maximized was the best approach. Table 1 demonstrates how sensitivity and specificity were used to determine how much money spent monthly best differentiated between recreational and 'at risk' gamblers in a sample of gamblers.

After determining suitable threshold for the gambling intensity variables we examined (including dollars spent annually, time spent per session, percent of annual income spent on gambling, and frequency of gambling) we found that exceeding each limit predicted harm from gambling (i.e., presence of two or more negative consequences) after controlling for known risk factors. We used logistic regression to accomplish this task. Logistic regression is used when predicting a binary outcome from other variables. Logistic regression produces odds ratios, which describe the likelihood of experiencing the outcome given the presence of other factors. We created risk curves from this information that demonstrated the relationship between gambling intensity and risk of experiencing negative consequences from gambling. For example, for dollars spent on gambling we found that the proportion of 'at risk' increased dramatically over the $85/month limit.
Results from these studies have been reported in papers, books, and at conferences:
Currie, S.R., Hodgins, D.C., & Wang, J. (2008). Canadian low-risk gambling limits: New evidence and limitations. Final report submitted to the Ontario Problem Gambling Research Centre, Guelph Ontario [report avaliable on OPGRC website, http://www.opgrc.org/contentdetail.sz?cid=3573&pageid=2147&r=s]
Currie, S. (2008, April). What is low risk gambling? Presented to Addictions Day Conference (Addiction Day: Broadening the Concept of Addiction), sponsored by the Calgary Health Region, Calgary, Alberta.
Currie, S.R., Miller, N.V., Hodgins, D.C., & Wang, J. (in press). Defining a threshold of harm from gambling for population health surveillance research. International Gambling Studies.
Currie, S.R., Wilhelm, A., Miller, N.V., Hodgins, D.C., & Wang, J. (2008, November). Predicting harm from gambling type clusters. Poster presented at Alberta Mental Health Board's 4th Annual Mental Health Research Showcase, Banff, AB. 
Click here for Poster
 
Wilhelm, A., Miller, N.V., Currie, S.R., Hodgins, D.C., & Wang, J. (2008, November). The impact of missing data in population surveys on problem gambling research in Canada. Poster presented at Alberta Mental Health Board's 4th Annual Mental Health Research Showcase, Banff, AB.
Click here for Poster
 
Currie, S.R., Hodgins, D.C., Wang, J., el-Guebaly, N., & Wynne, H. (2008). Replication of low-risk gambling limits using Canadian provincial gambling prevalence data. Journal of Gambling Studies, 24, 321-335. [Full article can be obtained from http://www.springerlink.com/content/1050-5350].
Currie, S.R. & Casey, D.M. (2007). Quantification and dimensionalization of gambling behaviour. In G.J. Smith, D.C. Hodgins, & R.J. Williams (Eds), Research and Measurement Issues in Gambling Studies (pp. 156-173). New York: Elsevier Inc.
Currie, S.R., Hodgins, D.C., Wang, J., el-Guebaly, N., Wynne, H., & Chen, S. (2006). Risk of harm from gambling in the general population as a function of level of participation in gambling activities. Addiction, 101, 570-580. [Full article can be obtained from http://www.blackwellpublishing.com ]
Currie, S.R., Hodgins, D.C., Wang, J., & Cunningham, J. (2007, September). Towards the development of Empirically-Based Responsible Gambling Limits. Paper presented at the Canadian Public Health Association Annual Conference, Ottawa, ON.
Currie, S.R., Hodgins, D.C., Wang, J., el-Guebaly, N., & Wynne, H. (2006, November). Cross validation of nationally-derived low-risk gambling limits with Alberta, British Columbia, and Ontario Gambling Prevalence Data. Paper presented at AMHB Mental Health Research Showcase, Banff, Ab.
Currie, S.R., Hodgins, D.C., Wang, J., el-Guebaly, N., Wynne, H., & Chen, S. (2005, November). Risk of harm from gambling in the general population as a function of level of participation in gambling activities. Poster presentation at AMHB Mental Health Research Showcase, Banff, AB.
 
Related Work by Other Research Groups
Independently, Jeremiah Weinstock and colleagues at the University of Connecticut employed the identical statistical methodology to identify quantitative limits of moderate gambling in pathological gamblers who continued to gamble following treatment. Harm form gambling was defined on a basis of reporting at least one symptom from the South Oaks Gambling Screen (SOGS). Among 178 pathological gamblers who participated in a cognitive-behavioral treatment program, gambling activity that reliably differentiated problem-free (SOGS score = 0) and symptomatic gambling (SOGS score equal to or less than 1) were gambling no more than once per month (sensitivity = 86%; specificity = 80%), gambling for no more than 1.5 hours per month (sensitivity = 83%; specificity = 87%), and spending no more than 1.9% of monthly income on gambling (sensitivity = 86%; specificity = 80%). This study is noteworthy because it suggests the concept of a quantitative limit may also apply to problem gamblers who choose controlled gambling rather than abstinence. This research group recently extended this approach to identify behavioral indicators of harmful gambling levels in college students (Weinstock et al., in press).
Weinstock, J., Ledgerwood, D.M., & Petry, N.M. (2007). Association between posttreatment gambling behavior and harm in pathological gamblers. Psychology of Addictive Behaviors, 21, 185-193. [Full article can be obtained from http://www.apa.org/journals/adb/ ]
Weinstock, J., Whelan, J.P., & Meyers, A.W. (2008). Gambling behavior of college students: When does it become harmful? Journal of American College Health, 56, 513-522.
Consultation with Other Researchers
Having determined quantitative limits for safe levels of gambling, the question remained whether these limits would be useful for public use. Our team consulted 171 gambling experts in North America to answer the question. These experts included both gambling researchers and clinicians treating gambling problems. The majority of the experts agreed that quantitative responsible gambling limits were important. The experts reported that percent of income spent on gambling was the most important quantitative responsible gambling limit, followed by frequency of gambling, amount spent gambling, and lastly duration of gambling. Surveyed experts also thought the limits suggested by our team were appropriate (i.e., not overly conservative or too liberal). Two key concerns voiced by the experts were: a) quantitative limits may promote a false sense of security for the public and b) problem gamblers may use the limits to justify gambling. Future research needs to investigate the impact of quantitative limits on vulnerable populations, such as 'at risk' gamblers or people from low income brackets.
Currie, S.R., Hodgins, D.C., Wang, J., el-Guebaly, N., & Wynne, H. (2008). In pursuit of empirically derived low-risk gambling limits. International Gambling Studies, 8, 207-227.
A panel discussion with 11 researchers from the fields of gambling and alcohol research was conveined on Sept. 3, 2008 to reach a consensus on methodological issues inherant in low-risk limits research. The results of this panel discussion are avaliable in the report below.
Currie, S.R., Hodgins, D.C., & Wang, J. (2008). Canadian low-risk gambling limits: New evidence and limitations. Final report submitted to the Ontario Problem Gambling Research Centre, Guelph, Ontario. [Report avaliable on OPGRC web site, http://www.opgrc.org/contentdetail.sz?cid=3573&pageid=2147&r=s

Social Media