1. Introduction

A current focus of gambling research is exploring gambling-related harms and identifying their range of impact (Brown et al., 2021; Lang- ham et al., 2016; Lind et al., 2022). Gambling-related harms can be experienced when excessive time or money is spent on gambling and are described as ‘adverse consequence due to an engagement with gambling that leads to a decrement in the health and wellbeing of an individual, family unit, community or population’ (Langham et al., 2016, p. 4). Harms occur across multiple domains, including financial and rela- tionship problems, emotional impacts, impacts to health, work and study, and involvement in criminal activity (Langham et al., 2016). Financial harms are often the first consequence of excessive gambling losses and can have an immediate impact, as well as having a cascading effect on other harms such as relational and emotional problems (Langham et al., 2016; Mathews & Volberg, 2013). These harms can affect individuals who gamble and those close to them; often termed “concerned significant others” (Concerned Significant Others; Castr´en et al., 2021; Dowling et al., 2014; Lind et al., 2022; Riley et al., 2018).

Gambling-related harms, by definition, are associated with decre- ments to the health and wellbeing of Concerned Significant Others. Commonly reported impacts on health and wellbeing are emotional or psychological distress (Castr´en et al., 2021; Chan et al., 2016; Dowling et al., 2016; A. Salonen et al., 2016; Svensson et al., 2013), symptoms of depression or mood disorders (Dannon et al., 2006; Wenzel et al., 2008), feelings of anger and guilt (Lorenz & Yaffee, 1988), alcohol or substance abuse (Svensson et al., 2013; Wenzel et al., 2008), poor self-assessed health (Chan et al., 2016), physical symptoms associated with anxiety (Lorenz & Yaffee, 1988) and reduced subjective wellbeing (Tulloch, Hing, et al., 2021). However, research on CSO health and wellbeing is primarily conducted with treatment-related populations, or populations who identify as being harmed by another person’s problem gambling (Tulloch, Browne, et al., 2021). The study populations used in the aforementioned papers are generally associated with respondents who are close to people with more severe gambling problems. Thus, it is currently unclear if these harms are also experienced by those close to gamblers experiencing a less severe gambling problem.

A public health approach stresses the importance of measuring the harms associated with gambling across the entire spectrum of gamblers (Korn et al., 2003). Researchers such as Canale et al. (2016) and Browne and Rockloff (2018) found that, in those who gamble, harms are not limited to those with a severe problem. Approximately 6.8% of the Australian adult population are categorised as at-risk (i.e. categorised as low- or moderate-risk gamblers by the PGSI [Ferris & Wynne, 2001]), compared to around 1.1% categorised as problem gamblers (Armstrong & Carroll, 2017). If only 1 in 5 of the at-risk individuals experienced harm, then this group would account for a larger proportion of harmed gamblers in the population than problem gamblers. Indeed, Browne and Rockloff (2018) found that, except for some very severe and rare harms, a greater proportion of the population aggregate harms could be attributed to at-risk gamblers than to problem gamblers. The harms experienced by people with less severe gambling problems are also associated with a variety of health, wellbeing and quality of life decre-ments; although these tend to be less severe than those experienced by problem gamblers (Blackman et al., 2019; Hilbrecht & Mock, 2019; Moayeri, 2020). For example, using disability weights, Browne et al. (2017) calculated the aggregate impact of gambling-related harms on quality of life, finding that while the severity of impact was greatest in those categorised as problem gamblers, at-risk gamblers also experi- enced some significant quality of life decrements. Overall, while the health and wellbeing impacts associated with at-risk gamblers are more clearly understood, little is known about Concerned Significant Others outside those associated with more severe gambling problems. It is possible that wellbeing im- pacts might also be found in people who have close relationships with at- risk gamblers. The number of Concerned Significant Others impacted by each problem gambler has been estimated at around six, and at-risk gamblers at about three (Goodwin et al., 2017). Despite the lower number of Concerned Significant Others impacted by at-risk gamblers, the higher prevalence of at-risk (6.8%) relative to problem gamblers (1.1%) implies a greater number of people potentially exposed. These people may be experiencing health and wellbeing decrements caused, or aggravated by, gambling-related harms occurring at subclinical levels. Increasing our knowledge of the impacts on Concerned Significant Others across the spectrum of gambling risk levels will inform and assist poli- cymakers and researchers in understanding the severity of gambling problems associated with the measurable impact on Concerned Significant Others’ health and wellbeing at the population level. In addition, this would allow for better targeting of policy for the greatest harm reduction benefits for both the person who gambles and those close to them.

1.1. Aims and objectives

The current study uses secondary analysis of a population- representative Australian survey. It examines the health and wellbeing of people living in households with others experiencing different se- verities of gambling problems. That is, people living in the same household as others categorised by the PGSI as ‘non-problem,’ ‘low- risk’, ‘moderate-risk’ and ‘problem’ gamblers. Specifically, the paper will extend existing research by aiming to understand if measurable health and wellbeing impacts are limited to only people living in the same households as gamblers with more severe problems (i.e., categorised as problem gamblers), or if they extend to those living in at-risk gamblers’ households (i.e., others categorised as low- or moderate-risk gamblers).

2. Methods

2.1. Study design and participants

The study analyses data from Wave 18 of the Household Income and Labour Dynamics in Australia Survey (HILDA; Department of Social Services & Melbourne Institute of Applied Economic and Social Research, 2019), the most recent wave to include a problem gambling screen. HILDA began in 2001 and is an ongoing Australian longitudinal survey collecting social and economic information. Full details are presented elsewhere (Summerfield et al., 2019; Watson & Wooden, 2012). In summary, Wave 1 began with a large national probability sample of 7,682 Australian households, collecting data from all in- dividuals within the household. Sample selection was via a stratified three-stage cluster design, covering all Australian households (except very remote; 0.8%; Australian Institute of Health and Welfare, 2019). The original sample was extended to include new household members as household compositions changed, and in Wave 11, it was topped up with an extra 2,153 households. In most cases, interviews were conducted face-to-face, except for gambling-related questions, which participants completed privately via a paper-based questionnaire. The 2018 HILDA

sample comprised 23,237 participants. Of these, participants did not fit eligibility for the current study if they were under 15 years of age (N = 4,831), they or their other household members had not completed the gambling-related questions (N = 3,679), or they had a gambling prob-

lem themselves (N = 1,029). Of those who did not complete the

gambling related questions, approximately 34% did not participate in the survey that year (non-responding person), 53% did not complete the Self Completion Questionnaire (containing, amongst others, gambling questions), and 13% refused or did not respond to the gambling ques- tions. The final sample of interest comprised 13,698 participants, who resided across 7,852 households (54.1% female, ages ranged between 15 and 84, mean age 46.3 years). Excluding children, respondents excluded

due to lack of gambling variables were significantly younger (m = 42.2 years; f(2,18403) = 68.97, p <.001) than either those excluded due to their own gambling problems (m = 46.6 years) or the included group (m

= 46.3 years), however, there was no significant gender difference (Х²(2) = 6.032, p =.51).

2.2. Measures

HILDA collects a broad range of measures; those relevant to this study are briefly described. A range of socio-demographic characteris- tics were assessed with questions probing age, gender, marital status, education, employment, and household income and debt. Problem gambling severity was assessed via the Problem Gambling Severity Index (PGSI; Ferris & Wynne, 2001). The PGSI is well-validated (Cron-bach’s alpha = 0.84; Ferris & Wynne, 2001) and consists of nine ques-tions used to measure gambling problems and identify symptoms of potentially harmful gambling behaviours. Total summed scores range between 0 and 27, and from these respondents are classified as either ‘non-problem gamblers’ (total score of 0), ‘low-risk gamblers’ (scores of 1 or 2), ‘moderate-risk gamblers’ (scores of 3 to 7), or ‘problem gam-blers’ (scores 8 + ).

Household categorisation was derived from the PGSI category of the person with the most severe gambling problem in the household, with each household classified as either a ‘non-problem gambling household’, a ‘low-risk gambling household’, a ‘moderate-risk gambling household’ or a ‘problem gambling household’. For example, if a household included a person categorised as a ‘problem gambler’ and one classified as a ‘moderate-risk gambler’, that household was categorised as a ‘problem gambling household’. Similarly, if a ‘moderate risk gambler’ and two ‘non-problem gamblers’ lived in the same household, this was categorised as a ‘moderate-risk gambling household’. Within these households, the study looked exclusively at people who did not have any level of gambling problem themselves (i.e., only those who

scored ‘0^′ on the PGSI). This enabled the comparison of only people who did not have personal gambling problems across different levels of gambling-risk households. All participants were asked to complete the

PGSI, as such, scores of ‘0^′ reflect both non-gamblers and non-problem gamblers, as we were not able to distinguish between the two.

HRQoL was assessed via the SF-36 (Ware & Sherbourne, 1992), a 36- item measure of functional health and wellbeing, validated for use in Australian populations (Butterworth & Crosier, 2004). All items (except for one asking respondents about health changes over the past year) are aggregated into eight multi-item scales and then transformed into a 0–100 scale with 0 representing the worst health status and 100 the best. The scales measure physical functioning, role-physical (interference with work or other daily activities due to physical health), bodily pain, general health, vitality, social functioning (interference with normal social activity), role-emotion (interference with work or other daily activities due to emotional problems), and mental health (symptoms associated with anxiety and depression and, conversely, measures of positive affect). Australian norms are provided by the Australian Bureau of Statistics (1995). Health state utility was measured by the SF-6D. This health index health state score is derived from the SF-36 and ranges from 0 (worst health state) to 1 (best health state). The SF-6D has been shown to have good psychometric characteristics (Baxter et al., 2015; Norman et al., 2013). HILDA calculates these utility values using both the UK and Australian weights (Norman et al., 2013; Walters & Brazier, 2003).

Life satisfaction was assessed using a single-item question that asked respondents, “All things considered, how satisfied are you with your life?” to gauge their global wellbeing. Responses are rated on an 11-point scale from 0 (totally dissatisfied) to 10 (totally satisfied). Do- mains of wellbeing were assessed via questions that asked people to rate how “satisfied or dissatisfied you are with some of the things happening in your life”. Participants were then shown a list which covered “Your employment opportunities”, “Your financial situation”, “The amount of free time you have”, “The home in which you live”, “Feeling part of your local community”, “The neighbourhood in which you live”, “How safe you feel” and “Your health”. Satisfaction with a partner was assessed via the question, “How satisfied are you with your relationship with your partner?”. Participants responded on an 11-point scale from 0 (totally dissatisfied) to 10 (totally satisfied). Most re- spondents reported within the “satisfied” range of the scale, which is typical of these measures (OECD, 2013).

2.3. Statistical analysis

The initial analyses use bi-variate statistical techniques to describe and explore the population utilised in the study (individuals with no gambling problem of their own) by the level of gambling risk experi- enced by others in their household. The SF-36 and life satisfaction scores are not normally distributed, having moderate negative skews ranging from -1.69 to -0.50. However, previous analysis suggests that, particularly in larger samples, the use of parametric statistical tech-niques with a standard normal error distribution are relatively robust and yield accurate results (Torrance et al., 2009; Walters & Campbell, 2005). These techniques have previously been used with HILDA data (e. g., Livingston, 2009; Renzaho et al., 2010). Accordingly, ANOVA was used to assess group differences between participants across each gambling household category for all SF-36 and life satisfaction domains. Welch was used where noted where the assumption of variance was violated, and Turkey’s HSD was used to assess differences between more than two groups. Following the bi-variate analyses, a series of multiple regressions were used to further examine the relationship between health and wellbeing and gambling problems in the household. The regressions continued to use the same population of interest; partici- pants with no gambling problem of their own (i.e., a PGSI score of 0). The level of ‘gambling household risk’ was included as a predictor in the regressions, as were some socio-demographic control variables commonly associated with health and wellbeing (gender, age, health, income, employment, and marital status). This model enabled a greater understanding of the unique impact the level of gambling problems in the household had on the health and wellbeing of the participant. Linearity was assessed via plots of studentized residuals against the predicted values. There was independence of residuals, as assessed by the Durbin-Watson statistic and no evidence of multicollinearity, assessed by tolerance values greater than 0.1. Missing data were removed listwise. Given the use of multiple tests in the study, we applied a Bonferroni correction and used adjusted alpha levels for our interpretations.

2.4. Ethics

The HILDA study was approved by the Human Research Ethics Committee of The University of Melbourne, and informed consent was obtained for all participants. In addition, approval for secondary anal- ysis was granted by Central Queensland University Human Research Ethics Committee (#22878).

3. Results

Participants in this study comprise only those categorised as non- problem gamblers (including non-gamblers) by the PGSI. Of the 13,698 participants in this study, 93.7% resided in a non-problem gambling household, 3.5% in low-risk gambling households, 1.9% in moderate-risk gambling households, and 0.9% in problem gambling households. Some demographic characteristics of these groups are shown in Table 1. Those living in problem gambling households tend to be significantly younger than in other groups, and those in low-risk and problem gambling households were more likely to be female. Those in non-problem gambling households were less likely to have never mar- ried and more likely to be widowed. They were more likely to have completed further education but had lower household incomes.

Fig. 1 illustrates HRQoL across each level of household gambling. There was a significant difference across gambling risk groups, F (3,13124) = 7.40, p <.001. Post-hoc tests revealed participants living in problem gambling households reported significantly lower SF-6D health

states than those in non-problem gambling households.

Table 2 details the domains of SF-36 HRQoL measures, for partici- pants across each of the household gambling risk levels, and Fig. 2 displays clustered means and error bars for these variables. There was a significant difference across gambling risk groups in the mental health domain. Post-hoc tests indicated that mental health was significantly poorer for people in problem gambling households than those in non- problem gambling households. No statistically significant differences were found across the remaining domains.

Table 3 outlines domains of satisfaction across each level of house- hold gambling risk and clustered means and errors bars are displayed in Fig. 3. There were significant differences across gambling risk groups in satisfaction with financial situation and feeling part of your local com- munity. Post-hoc tests showed participants living in problem gambling households reported significantly lower mean satisfaction scores than those in non-problem gambling households in two domains: their financial situation and feeling part of their local community. Significantly lower satisfaction levels with their financial situation were also found in participants in moderate-risk gambling households compared to those in non-problem gambling households. No other significant differences were found.

Multiple regressions were conducted to control for non-gambling factors commonly associated with HRQoL (gender, age, health condi- tions, marital status, income, education, and employment). As shown in Table 4, compared to living in a non-problem gambling household, living in problem gambling households significantly predicted SF-6D and satisfaction with feeling part of the local community, but not the SF-36 mental health scale. After controlling for demographic variables, living in any at-risk or problem gambling household all significantly depression. These mean differences were within the range to be considered clinically and socially meaningful (Ware et al. 1994; cited in Butterworth & Crosier, 2004; Walters & Brazier, 2003). Unlike results that have been found in people with their own gambling problems illustrated by the error bars in Fig. 1, it may be that some differences are not detectable given the available sample size, and within-group vari- ability caused by the number of other contributors to health and well- being. The magnitude of the effect of living in problem gambling households on HRQoL is approximately two to three times greater than low-risk or moderate \-risk households. This ratio is similar to that found

in people who gamble (Browne et al., 2017). Overall, these findings may reflect that living in households with severe, but not necessarily low to moderate gambling problems, contributes to ill-health via factors iden- tified in previous research. These include increased stress and guilt, in turn leading to poorer sleep (Landon et al., 2018) and headache, bowel and stomach issues (Lorenz & Yaffee, 1988). Serious gambling problems may be specifically associated with unhealthy lifestyles, such as alcohol and tobacco use by both gamblers and Concerned Significant Others (Morisano et al., 2009; A. H. Salonen et al., 2015; Svensson et al., 2013).

Concerning subjective feelings of wellbeing, the study found signif- icantly lower satisfaction with their financial situation was experienced in moderate-risk and problem gambling households, with a larger effect in problem-gambling households. This is despite these groups having significantly higher household incomes than non-problem gambling households, and no significant difference in household debt. This finding supports the understanding around gambling harms as detailed in Langham et al.’s conceptual framework (2016), with financial problems most directly related to gambling and often appearing in people expe- riencing less severe gambling problems. A CSO’s dissatisfaction with their financial situation can then contribute to increased stress as well as predicted lower satisfaction with their financial situation, with the financial effects greater in those living in problem gambling households.

4. Discussion

To our knowledge, this is the first article to explore the relationships between the severity of gambling problems and the health and wellbeing of others within the household. We aimed to understand if measurable health and wellbeing impacts were limited only to people living in the same households as gamblers with severe problems or if they extended to those living in at-risk gamblers’ households. This enabled us to extend existing research by exploring the impacts of gambling-related harms outside treatment-seeking or problem gambling populations. Addition- ally, by examining the health and wellbeing of only people who were either non-problem gamblers or non-gamblers, the study effectively controlled for own-gambling problems, which are relatively common in Concerned Significant Others (Mazar et al., 2018; Meisel et al., 2013).

The study found a consistent linear trend for lower health and wellbeing associated with increased gambling problems in the house- hold. With HRQoL, living in a problem gambling household was asso- ciated with significantly worse health states than living in non-problem gambling households. Participants living in a problem gambling household also reported significantly worse mental health functioning in the SF-36, a sub-scale that measures feelings of nervousness and/or relational and psychological health issues (e.g., Ja¨rvinen-Tassopoulos, 2020). Browne and Rockloff (2018), found relatively serious financial harms are commonly experienced by low- and moderate-risk gamblers, but that more severe health and social impacts were largely limited to problem gamblers. Finally, financial harms can have a long-term impact on those experiencing them (‘legacy’ harms; Langham et al., 2016). As such, this problem may continue to affect the Concerned Significant Others long after the gambling problem is resolved or they have left the household, and thus may have a long-term impact on a CSO’s wellbeing.

Living in a problem gambling household was also significantly associated with lower satisfaction with feeling part of the local com- munity. The scores and mean differences reported in ‘feeling part of the community’ are similar to those reported between people experiencing very high psychological distress compared to those who are not (Reeve et al., 2016) and those currently experiencing material deprivation versus not (The Melbourne Institute, 2020). High levels of preoccupa- tion with gambling, or dealing with the issues caused by another’s gambling, may act to supplant other activity, and dissolve bonds with one’s community/neighbourhood. Feelings of shame, stigma and guilt, often experienced by Concerned Significant Others (Ja¨rvinen-Tassopoulos, 2020), may also be causing a barrier between themselves and those around them. The combined findings of dissatisfaction with finances and dissatisfaction with feeling part of the local community have been associated with a transition into loneliness (Baker, 2012), which has its own serious health and wellbeing implications (Leigh-Hunt et al., 2017).

4.1. Limitations

The results of this study should be interpreted in light of several limitations. First, caution must be taken in the overall interpretation of these results due to the limitations regarding power. Serious gambling problems are relatively uncommon in general population samples; therefore, the study may lack the statistical power necessary to detect some small effects. Health and wellbeing are impacted by a wide range of factors including societal and personal characteristics, genetic and demographic variables (Dolan et al., 2008; Geerling & Diener, 2020), so any single contributor tends to exert a small impact, relative to total variation. For this reason, future work would benefit from using non- population representative stratified samples and attempting to control for as many other influences on HRQoL as possible, to reduce unex- plained variance. No significant differences were identified between either category of at-risk households (low- or moderate-risk) and non- problem gambling households but this may be (again) due to issues of low power. Bias may have been introduced into the study, due to the using particular sub-sample of HILDA respondents (whose household completed the PGSI and who didn’t have a gambling problem them- selves). However, there were no gender differences and only a small age difference between the included and excluded groups, and it was un- feasible to include participants where the entire household’s gambling risk levels were unknown. Finally, the bi-directional nature of the relationship between gambling problems and health and wellbeing means that while associations were found, both causal directions are possible.

4.2. Conclusions

The overall results seem to indicate a step-change in the health and wellbeing impacts of harm to Concerned Significant Others, between moderate-risk and problem gambling households. At the population level, it appears that it is at the more severe end of the gambling-risk spectrum that a CSO (and not experiencing gambling problems of their own) starts to experience measurable negative impacts on their wellbeing. As an exception, some financial impacts on wellbeing appear to be associated with less severe problems. Concerned Significant Others are often unaware of gambling problems until they arequite serious (Kourgiantakis et al., 2013; Riley et al., 2018), so Concerned Significant Others may experience dissatisfaction with their financial situation before they are aware that gambling may be causing those financial issues within the household. Dissatisfaction in this area could be explored as a way to identify Concerned Significant Others associated with less severe gambling problems in order to offer support. Additionally, financial counselling might be useful for both moderate risk and problem gamblers to support both them and their families.

Without limiting the impact of the harms experienced by Concerned Significant Others of at- risk gamblers, it does appear that the bulk of the measurable effects on health and wellbeing found in this study was experienced by people living in households where others are suffering from more severe gambling problems. Policy considerations and investments should focus on strategies that prevent gambling harms in order to reduce the health and wellbeing burden across the entire community, while also targeting help and assistance to those close to people with severe gambling problems.