Introduction

Gambling is associated with various degrees of harm and problematic gambling is considered a public health issue in many countries. Harm includes, in addition to obvious financial negative impact, relationship problems, emotional or psychological distress, impaired health, cultural harm, impaired academic or work performance as well as criminal activity (Langham et al., 2016). At the extreme, gambling disorder is the formal medical diagnosis that is characterized by impairment of control, continued gambling despite significant negative consequences and increasing priority given to gambling over other activities (American Psychiatric Association, 2013).

There is a common interest for society, the gambling industry and individuals and their families to prevent gambling problems. Prevention requires a multifaceted approach (Christensen, 2020; Hing et al., 2019) focusing on the parameters of the gambling products and environment as well as individual-level initiatives. Prevention can take place at the primary, secondary and tertiary level (Caplan, 1964). In the alcohol field, a primary prevention measure since the 1990s has been to develop recommendations for “low-risk drinking”, advice to consumers on the maximum number of standard drinks a person can drink per day and per week and still have a relatively low risk of developing negative, typically physical health, consequences. Inspired by this, attempts have been made to develop similar recommendations for gambling (Brosowski et al., 2015; Currie et al., 2009, 2012, 2017; Dowling et al., 2018). Unlike alcohol, where physical harm can be defined and measured reliably, gambling-related harm is multidimensional and more challenging to measure. Moreover, whereas it is assumed that people of the same sex and body weight are affected fairly equally by the physical harmful effects of alcohol, the harmful effects of gambling are partially determined by how much time and money a person has at their disposal. A person with very scarce financial circumstances can experience negative gambling-related consequences at lower losses than one with more financial resources.

Solutions to some of the challenges associated with identifying lower risk gambling guidelines have evolved over the course of a number of studies by different investigator groups (Brosowski et al., 2015; Currie et al., 2012; Dowling et al., 2021). For example, one way to handle individual differences in financial means is to examine gambling expenditure as a proportion of the individual's income versus simple expenditure (Currie et al., 2017). Using actual player data from gambling accounts is another approach with unique strengths and limitations (Gainsbury, 2011). Two studies have addressed the issue of the accuracy of gambling behaviour as measured in population surveys by using actual recorded player data versus retrospectively self-reported gambling. In the first of these studies, Brosowski, Meyer, and Hayer, (2012) applied lower risk guidelines calculated from Canadian survey data collected in the same timeframe (Currie et al., 2006) to player data from BWin, a European online gambling provider. They reported that involvement in a larger number of gambling formats increased the likelihood of exceeding the limits, and that exceeding the limits was associated with increased likelihood of players choosing to close their accounts due to gambling-related problems. More recently, Louderback, LaPlante, Currie, and Nelson (2021), also using data from BWin, compared the performance of lower risk guidelines calculated from a Canadian longitudinal survey sample (Currie et al., 2017) to limits they calculated from online gambling data. The latter limits were higher, and, for some harm indicators, more strongly predictive. They concluded that online gambling might require different limits than land-based gambling because of its higher risk profile (Brosowski, Olason, Turowski, & Hayer, 2021; Gainsbury, 2015).

The Louderback et al. (2021) study had a number of limitations. The authors noted that their study focused on predicting risk of gambling disorder and not harms conceptualized more broadly as in other studies. Individuals not meeting the criteria for gambling disorder experience significant amounts of harm (Langham et al., 2016). Another limitation is that the entirety of individuals' gambling was not available for analysis. Many online gamblers also gamble on land-based venues (Lind et al., 2021). Although the participants' Bwin gambling was reliably assessed, land-based and gambling on other online sites were not included. In the current study we use player data that reflects both online and venue-based gambling, capturing the majority of individuals' gambling involvement by accessing data from a Norwegian monopoly gambling company.

Method

Results

Risk curve analysis

Frequency

As seen in Fig. 1, a high proportion is reporting harm even at a low frequency of gambling and a linear increase of harm is found from a frequency around 9 days, forming a J-shaped curve. As seen in Table 1, AUCs for the four harm indicators ranged from 0.55 to 0.59 and 8.68 days per month was identified as the optimal cut-off across all four indicators. Optimal cut-off for 18–25 year olds was 5–5.67 days. The proportion of people reporting harm reported below the cut-off is between 11% and 18% and the increase in risk above the cut-off is between 1.4 and 1.8 times across the four harm categories.

View Full Size

Risk curves for frequency, expenditure, duration, gambling formats and wager on Two plus harm and financial harm, social harm and emotional harm

Citation: Journal of Behavioral Addictions 11, 3; 10.1556/2006.2022.00062

• Download Figure
• Download figure as PowerPoint slide

Table 1.

Gambling related harm categories and ROC results for gambling frequency, expenditure, duration, gambling formats and wager

Gambling frequency	ROC results			OR	% harm below cut-off	% harm above cut-off
	AUC (95% CI)	Sensitivity/Specificity	Optimal cut-off (18–25 year old)
Financial harm	0.55 (0.56–0.58)	0.37/72	8.68 days (5.67 days)	1.45	13.4	19.5
Social harm	0.59 (0.58–0.60)	0.42/73	8.68 days (5.67 days)	1.77	11.2	19.9
Emotional harm	0.59 (0.58–0.60)	0.42/73	8.68 days (5.67 days)	1.74	13.4	23.3
2+ harms	0.59 (0.58–0.59)	0.40/0.74	8.68 days (5 days)	1.63	18.0	29.4

Expenditure	ROC results			OR	% harm below cut-off	% harm above cut-off
	AUC (95% CI)	Sensitivity/Specificity	Optimal cut-off (18–25 year old)
Financial harm	0.55 (0.54–0.56)	0.40/0.72	53.8 € (32.6 €)	1.57	12.9	20.3
Social harm	0.57 (0.56–0.58)	0.43/72	53.8 € (32.6 €)	1.77	11.2	19.8
Emotional harm	0.58 (0.57–0.59)	0.44/0.73	53.8 € (32.6 €)	1.80	13.1	23.6
2+ harms	0.57 (0.56–0.58)	0.42/0.73	53.8 € (32.6 €)	1.66	17.8	29.6

Duration	ROC results			OR	% harm below cut-off	% harm above cut-off
	AUC (95% CI)	Sensitivity/Specificity	Optimal cut-off (18–25 year old)
Financial harm	0.63 (0.62–0.64)	0.50/0.73	>83 min (57 min)	2.29	10.9	25.0
Social harm	0.66 (0.65–0.67)	0.54/0.74	>83 min (57 min)	2.68	9.1	24.5
Emotional harm	0.66 (0.65–0.67)	0.54/0.74	>83 min (57 min)	2.68	10.7	29.0
2+ harms	0.66 (0.65–0.67)	0.56/0.71	>72 min (45 min)	2.36	14.8	36.4

Number gaming formats	ROC results			OR	% harm below cut-off	% harm above cut-off
	AUC (95% CI)	Sensitivity/Specificity	Optimal cut-off (18–25 year old)
Financial harm	0.64 (0.63–0.65)	0.27/0.90	<3 (<3)	2.54	12.7	32.2
Social harm	0.65 (0.64–0.66)	0.28/0.90	<3 (<3)	2.63	11.4	30.0
Emotional harm	0.66 (0.65–0.67)	0.28/0.90	<3 (<3)	2.68	13.4	35.9
2+ harms	0.66 (0.65–0.67)	0.27/0.91	<3 (<3)	2.50	17.9	44.8

Wager	ROC results			OR	% harm below cut-off	% harm above cut-off
	AUC (95% CI)	Sensitivity/Specificity	Optimal cut-off (18–25 year old)
Financial harm	0.64 (63–0.64)	0.51/0.74	139.6 € (81.5 €)	2.40	10.7	25.6
Social harm	0.66 (0.66–0.67)	0.55/0.74	139.6 € (81.5 €)	2.84	8.9	25.2
Emotional harm	0.68 (0.67–0.68)	0.55/0.75	139.6 € (81.5 €)	2.87	10.4	29.9
2+ harms	0.66 (0.66–0.67)	0.57/0.71	118.2 € (81.5 €)	2.41	14.6	37.2

Expenditure

As seen in Fig. 1, the proportion of the sample reporting harm was quite high at all levels of expenditure. Higher harm was found among players winning and losing more than around 55 €, forming U-shaped curves. As seen in Table 1, AUC for the four harm indicators was between 0.55 and 0.58, with 53.8 €/month as the suggested optimal cut-off across all indicators. It corresponds to approximately 1% of median household gross income in Norway². Optimal cut-off for 18–25 year olds was 32.6 €. The percentage of the sample reporting harm who were below the cut-off ranges between 11% and 18% and the increase in risk above the cut-off is between 1.6 and 1.8 times across the four harm categories.

Duration

As seen in Fig. 1, the proportion of the sample reporting harm is above 10% at all duration levels. Starting flat, a linear increase is found from a duration just above one hour, forming J-shaped curves. As seen in Table 1, AUC for the four harm indicators was between 0.63 and 0.67, with 72 min as suggested optimal cut-off for Two plus harm. The cut-off for the GamTest dimensions was >83 min. Optimal cut-off for 18–25 year olds was 45–57 min. The proportion of the sample reporting harm below the cut-off was between 9% and 15% and the increase in risk above the cut-off is between 2.3 and 2.7 times across the four harm categories.

Number of gambling formats

As seen in Table 1, AUC for the four harm indicators ranged between 0.64 and 0.66, with more than two gambling forms as suggested cut-off. As seen in Fig. 1, there was a linear increase in risk of harm with an increased number of formats played. As seen in Table 1, the harm reported below the cut-off was between 11% and 18% and the increase in risk above the cut-off is between 2.5 and 2.7 times across the four harm categories.

Wager size

As seen in Fig. 1, the proportion of the sample reporting harm is around 10% at the lower wager sizes. Starting flat, a linear increase is found from a wager just above 100 € forming J-shaped curves. As seen in Table 1, AUC for the four harm indicators was between 0.64 and 0.68, and >118.2 € as suggested cut-off for Two plus harm. The optimal cut-off for the GamTest dimensions was >139.6 €. It corresponds to approximately 2.1%–2.5% of median household gross income in Norway. Optimal cut-off for 18–25 year olds was 81.5 €. The harm reported below the cut-off is between 9% and 15%, and the increase in risk above the cut-off is between 2.4 and 2.9 times.

Low risk limits and harm

As seen above, there are persons gambling below the low-risk limits who experience harm. Table 5 presents gender, age and gambling participation for four groups: below all low-risk limits and no harm, below all low-risk limits and harm, above any low-risk limit and no harm, above any low-risk limit and harm. Comparing the two below low-risk limit groups shows that the group experiencing harm are younger, have a higher proportion of men and a higher participation in betting and online casino than the group with no harm who have higher participation in lottery. For the two groups above low-risk limits, we see the same pattern and also a higher participation in VLT in the harm group.

Table 5.

Gender, age and gambling participation among four groups defined by gambling under/over low-risk level and experiencing no harm/harm

Discussion

Data from a specific gambling commercial environment, comprising actual gambling behaviour, were analyzed in a search for reliable and valid lower risk limits. GamTest (Jonsson et al., 2017) items were used as harm indicators for financial, social and economic harm.

In the ROC analyses, AUCs were just below 0.60 for Frequency and Expenditure and for Duration, Number of gambling formats and Wager AUC were around 0.65. Although AUC of 0.60 is considered adequate for longitudinal prediction (Currie et al., 2017), these AUC values are somewhat lower than previous research with data from prevalence studies, which is surprising given that the current objectively recorded gambling data include less measurement error than reported data. However, because only the individuals gambling with NT was captured, gambling on other platforms may also be contributing to gambling-related harms. In previous risk curve analyses using prevalence data, all gambling is normally included in an individual's self-report, albeit influenced by memory and other reporting biases.

The suggested low-risk limits in the present study are higher than limits suggested by the lower risk guidelines derived from international data sets (Young et al., 2022) and those identified by Currie et al. (2017). In contrast, the limits for frequency and wager are lower and net loss higher compared to Louderback et al. (2021) who also used data from a gambling company versus prevalence survey data. Their datasets are older, from between the years 2005 and 2011, which partly could explain the differences. In addition, Bwin and NT's have different returns to the player. As a monopoly company, NT has lower return to player than common offshore companies, partly due to product mix. This in turn yields higher net losses for the same wager.

The low-risk limits for 18–25 years old presented in this study are significantly lower than the low-risk limits based on the whole sample. This is unsurprising since younger people are more vulnerable to gambling problems and generally have fewer financial resources. This result indicates that one should consider special low-risk recommendations for younger persons.

One notable observation of these findings is that there is a significant number of gamblers experiencing harm in the reference groups which reflect gambling levels well below the suggested cut-offs. Those gambling below low-risk levels and experiencing harm tend to be younger and playing high-risk games to a greater extent than those playing below low-risk limits with no harm. In fact, even at the lowest levels of gambling, at least 5% and typically 10% of individuals report harm. This could partly be explained by the fact that many customers at NT also play at other companies. It is also a consequence of using a relative risk approach to cut-off development which identifies a threshold that optimizes the identification of harm versus no harm, above and below the cut-off. It does not determine the best cut-off that prevents all harm.

The shape of risk curves for gambling problems and their meaning have been subject to debate. The risk curves presented by Currie et al. (2009, 2012, 2017) for different gambling behaviours were J-shaped. Markham et al. (2016) found r-shaped risk curves for gambling frequency in three different countries and a linear risk curve in Norway. Using quintiles on the x-axis, Louderback et al. (2021) describe the risk curves as J-shaped when using online gambling data, but these could also be described as having a shape fairly close to linear. In the present study, the shapes of most risk curves show a rather stable harm level up to a certain point, from which the increase in harm is fairly linear. From a gambling company's perspective, this indicates that customers at-risk of harm can be found at all levels of consumption. This is contrary to the “Reno Model” (Blaszczynski, Ladouceur, & Shaffer, 2004) that recommends that social gamblers should not be subject to prevention measures that might disturb their play. In a gambling market where gambling through the internet has become increasingly available and most gamblers are active at several companies, minimizing harm is important for all customers.

Although the present study used player data from a large sample of NT customers, it still comes with some limitations. One limitation is that the respondents are self-selected as they had elected to take the GamTest to assess gambling impacts. The NT Playscan system provides messaging to all players, including invitations to self-assess problematic gambling. The GamTest is a validated measure of problematic gambling and is a face valid measure of harm, but self-completion measures are imperfect (Swets, Dawes, & Monahan, 2000). It should also be noted that the sample was restricted to customers of one of two monopolists in Norway, which further limits the generalizability of the findings. Another limitation is the low AUC values, probably due to the fact that an unknown number of customers also are playing at other sites than NT or because other variables than those assessed may be related to harm. The player data used stemmed from three months before the GamTest was completed, hence it cannot be ruled out that other time spans of data could have provided different results. In studies using self-report data, the questions on gambling activity asked people to estimate ‘typical' monthly consumption. To replicate this approach with actual gambling activity in this study, the mean over three months is used as a proxy for ‘typical’ consumption. Also, as with all ROC-analysis there will be some misclassifications, implying the cutoffs do not apply to all individuals. It should also be noted that fixing the sensitivity to a minimum of >0.70 could have resulted in somewhat lower cut-offs than allowing lower sensitivity values.

The results in the present study require replication and because predictive validity is limited, refinements in methodology are also important to capture accurate and complete gambling data. One possibility would be to ask the customers as part of the self-assessment about gambling with other gambling sites. Another more challenging way would be to ask the participants for permission to collect gambling data directly from all their sites they use. This would however require cooperation of numerous providers. Asking customers about their financial situation (e.g., household income) would provide the possibility to suggest low-risk limits regarding this as well as the possibility for the gambling companies to use it in their duty of care.

The lower risk gambling guidelines published by Young et al. (2022) incorporate the feedback and reactions to various draft versions of individuals who gamble, ranging from light to heavy gambling engagement. This type of validation research is also important with online gamblers. Moreover, assessing how individuals use this advice in monitoring and modifying their play is crucial.

In summary, the suggested lower-risk limits in the present study are higher than limits based on prevalence studies. There was a significant number of gamblers (5–10%) experiencing harm at gambling levels well below the suggested cut-offs and the risk increase at certain consumption levels. Customers at-risk of harm can be found at all levels of consumption. In an increasingly available gambling market where most gamblers are active at several companies, minimizing harm is important for all customers.

Funding sources

Funding for this project was provided by La Foundation Mise sur Toi (registered independent charity with funding from Lotto Québec) to the Canadian Centre on Substance Use and Addiction.

Authors’ contribution

Study concept and design: AL, DCH, MMY, JJ, SC. Analysis and interpretation of data: All authors made substantial contribution to the analysis and interpretation of findings. Statistical analysis: JJ. Obtained funding: JJ. All authors revised the manuscript for important intellectual content. JJ and AL had full access to all data in the study and take full responsibility for the integrity of the data. JJ take full responsibility for the accuracy of the statistical analysis.

Conflict of interest

JJ is an employee at Sustainable Interaction, a private company working with responsible gambling and online training with the gambling industry. He received consulting fees from the Canadian Centre on Substance Use and Addiction related to this project. He has also been working as a consultant for the Norwegian gambling operator Norsk Tipping among other gambling operators. He has formerly published research partly financed by Norsk Tipping. He has received research grants by the Svenska Spel research council during 2021. Svenska Spel is a state-owned Swedish gambling company and the research council, sovereign in their decisions, is funded by Svenska Spel. DCH receives partial salary support and competitive research funding from the Alberta Gambling Research Institute, a government funded research granting agency. He also received consulting fees from the Canadian Centre on Substance Use and Addiction related to this project. AL is employed by AB Svenska Spel and involved in the responsible gambling work of the company. He has also worked for Norsk Tipping through the by them licensed responsible gambling tool Playscan. SC received consulting fees from the Canadian Centre on Substance Use and Addiction related to this project. MMY reports no conflicts of interest. SP reports no conflict of interest. PC reports research funding by several gambling providers in Sweden, Finland and Norway, as well as ongoing academia-industry research collaborations on scientific evaluations of responsible gambling tools.