Introduction
Although gambling for recreation is common in many societies, some individuals develop problem gambling, which ‘disrupts or damages family, personal or other recreational pursuits' (Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011). Problem gambling is considered to lie on a continuum with pathological gambling (Shaffer & Martin, Reference Shaffer and Martin2011; Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011) and is classified as an impulse control disorder in DSM-IV (APA, 2000), with similarities to addiction disorders (Goudriaan et al. Reference Goudriaan, Oosterlaan, deBeurs and van den Brink2006; Shaffer & Martin, Reference Shaffer and Martin2011). Gambling problems are associated with significant individual and societal costs (Bowden-Jones & George, Reference Bowden-Jones and George2011). Although causal relationships are difficult to establish, the societal and economic costs of problem gambling include those related to incurred debt and household financial stress, reduced productivity, absenteeism and unemployment, divorce, domestic violence and other criminal behaviours (Reith, Reference Reith2006; Walker, Reference Walker2007; Victorian Competition and Efficiency Commission, 2012). Between five and 10 individuals including family members, friends, co-workers and family members are estimated to be adversely affected by each individual problem gambler (Reith, Reference Reith2006). Such costs are difficult to quantify and have been estimated as between US$9300 and US$53 000 per annum for each problem gambler (Walker, Reference Walker2007). However, it is recognized that these costs may be imprecise and underestimated and further research is required to quantify direct and indirect costs of problem gambling (Reith, Reference Reith2006; Walker, Reference Walker2007).
Population-based surveys have established that problem gamblers are more likely to occupy lower socio-economic positions in society (Gotestam & Johansson, Reference Gotestam and Johansson2003; Welte et al. Reference Welte, Barnes, Wieczorek, Tidwell and Parker2004; Kessler et al. Reference Kessler, Hwang, LaBrie, Petukhova, Sampson, Winters and Shaffer2008; Bakken et al. Reference Bakken, Gotestam, Grawe, Wenzel and Oren2009; Afifi et al. Reference Afifi, Cox, Martens, Sareen and Enns2010; Tavares et al. Reference Tavares, Carneiro, Sanches, Pinsky, Caetano, Zaleski and Laranjeira2010), have mental health problems (Petry et al. Reference Petry, Stinson and Grant2005; Kessler et al. Reference Kessler, Hwang, LaBrie, Petukhova, Sampson, Winters and Shaffer2008; Lorains et al. Reference Lorains, Cowlishaw and Thomas2011), experience disabilities (Morasco & Petry, Reference Morasco and Petry2006) and report alcohol and substance misuse (Welte et al. Reference Welte, Barnes, Wieczorek, Tidwell and Parker2004; Petry et al. Reference Petry, Stinson and Grant2005; Kessler et al. Reference Kessler, Hwang, LaBrie, Petukhova, Sampson, Winters and Shaffer2008; Lorains et al. Reference Lorains, Cowlishaw and Thomas2011; Shaffer & Martin, Reference Shaffer and Martin2011).
Mechanisms predisposing individuals to problem gambling are not well established but include a range of neurocognitive deficits and distortions (Goudriaan et al. Reference Goudriaan, Oosterlaan, deBeurs and van den Brink2004; Xian et al. Reference Xian, Shah, Phillips, Scherrer, Volberg and Eisen2008), such as greater impulsivity (Forbush et al. Reference Forbush, Shaw, Graeber, Hovick, Meyer, Moser, Bayless, Watson and Black2008), a preference for smaller but immediate rewards over larger delayed ones (Dixon et al. Reference Dixon, Marley and Jacobs2003) and an impaired ability to learn from previous experience (Goudriaan et al. Reference Goudriaan, Oosterlaan, deBeurs and van den Brink2005). Many of the characteristics associated with problem gambling are also observable in individuals with a lower general cognitive ability or intelligence, most commonly measured by IQ tests (Shamosh et al. Reference Shamosh, Deyoung, Green, Reis, Johnson, Conway, Engle, Braver and Gray2008; Lubinski, Reference Lubinski2009). It is therefore plausible that deficits in general cognitive ability (i.e. a lower IQ) may be associated with a higher risk of problem gambling. However, there is very little research testing this possibility, particularly in representative samples. Clarification of this issue may advance our understanding of the causes of problem gambling, and help to identify high-risk groups and plan protective approaches and targeted interventions. We therefore investigated correlates of problem gambling in a large population-based study in England, focusing on IQ while controlling for potentially important characteristics including socio-economic circumstances, impulsive and addictive behaviours, and mental health problems.
Method
The Adult Psychiatric Morbidity Survey (APMS) 2007
The APMS 2007 was designed to assess the mental health of a representative sample of the population aged ⩾16 years, living in private households in England. A multistage stratified probability sampling design was used. The Royal Mail's Small User Postcode Address File (PAF) was the sampling frame, with postcode sectors (comprising 2550 households on average) representing the primary sampling units. Postcode sectors were first stratified by Strategic Health Authorities (SHAs). All the primary sampling units within each SHA were then further stratified on the basis of the proportion of persons in non-manual classes, and sorted by the proportion of households without a car based on 2001 Census data. Postal sectors were then sampled from each stratum with a probability proportional to size, and 519 were thereby selected. Within these postal sectors, 28 delivery points were randomly chosen, yielding a total of 14 532 delivery points. Interviewers visited these addresses to identify private households with at least one person aged ⩾16 years. Within the potentially eligible sample of 12 694 addresses, one person from each household where contact was made was randomly chosen to take part in the survey, and 7461 (57%) individuals agreed to be interviewed (69% of those successfully contacted). The fieldwork was completed between October 2006 and December 2007. The interviews were carried out by experienced National Centre for Social Research (NatCen) interviewers, many of whom had worked previously on health-related surveys. They were briefed and trained on the administration of all modules of the survey and were provided with full sets of written instructions. As the fieldwork took place over the course of 1 year, interviewers who took a break from the survey for 4 months or more were provided with refresher sessions. A project supervisor accompanied all interviewers during the early stages of their fieldwork to ensure that the interviews were administered correctly. In subsequent interviews, routine supervision of 10% of interviewer work was carried out. Complete details of the survey methodology are available elsewhere (McManus et al. Reference McManus, Meltzer, Brugha, Bebbington and Jenkins2009).
Ascertainment of problem gambling
A questionnaire based on the 10 DSM-IV diagnostic criteria for pathological gambling (see Table 1) was administered using computer-assisted self-interviewing. Although there is no agreed gold standard instrument to measure pathological or problem gambling, the DSM-IV screening instrument was used so that the results could be compared with those of the British Gambling Prevalence Surveys (BGPS), which use a similar methodology (Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011). The questions related to gambling in the APMS were divided into two sections. The first section established whether the individual had spent any money on gambling in the previous year. Participants were given examples of gambling activities that included buying lottery tickets or scratch cards; playing games or making bets for money on the internet; playing football pools, bingo or fruit machines; playing games or making bets with friends for money; betting on races (with/without a bookmaker); and casino gaming. Participants were then asked: ‘Have you spent any money on any of these things in the last 12 months?’ Those who answered ‘yes’ were administered the problem gambling screen based on the 10 DSM-IV criteria. Those who answered ‘no’ were asked a check question about whether they had gambled just occasionally in the past year, such as buying a lottery ticket or scratch card. Individuals reporting occasional gambling in this additional question were also administered the problem gambling screen. The numbers of DSM-IV criteria endorsed in the screen were summed to generate a score (those who had not gambled in the past year were assigned a score of zero). As in the BGPS and other similar studies (Gotestam & Johansson, Reference Gotestam and Johansson2003; Bakken et al. Reference Bakken, Gotestam, Grawe, Wenzel and Oren2009; Afifi et al. Reference Afifi, Cox, Martens, Sareen and Enns2010; Brewer et al. Reference Brewer, Potenza and Desai2010; Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011), a score of ⩾3 was considered to indicate problem gambling. DSM-IV pathological gambling requires a score of ⩾5, but previous studies, including the BGPS, have generally identified too few cases for meaningful statistical analyses (Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011).
DSM-IV criteria for pathological gambling and the gambling questions included in the survey
Ascertainment of IQ
The National Adult Reading Test (NART), a widely used validated measure for English speakers, was used to measure verbal IQ (Nelson, Reference Nelson1991; Crawford et al. Reference Crawford, Deary, Starr and Whalley2001; Bright et al. Reference Bright, Jaldow and Kopelman2002). The NART comprises a list of 50 words that respondents read out. The interviewer counts and records the number of reading errors made. A validated algorithm was used to convert this error score to a revised Wechsler Adult Intelligence Scale (WAIS-R) verbal IQ estimate [predicted WAIS-R verbal IQ = 127.4 – (1.14 × NART error score)] (Nelson, Reference Nelson1991). Participants whose first language was not English, and those who refused to complete the NART, were not administered the test. Respondents who indicated that they could not read the words because of poor eyesight or other problems such as dyslexia were not included in the verbal IQ calculation.
The NART is highly correlated with full-scale IQ scores on the WAIS-R and other IQ measures in clinical and non-clinical samples, and is thus a good proxy of general intellectual ability, which is particularly appropriate for large-scale population-based studies (Nelson, Reference Nelson1991; Crawford et al. Reference Crawford, Deary, Starr and Whalley2001; Bright et al. Reference Bright, Jaldow and Kopelman2002; Mathias et al. Reference Mathias, Bowden and Barrett-Woodbridge2007; Spinks et al. Reference Spinks, McKirgan, Arndt, Caspers, Yucuis and Pfalzgraf2009). NART scores are also largely unaffected by the presence of mental illness or neurological disorders, making them ideal for a mental health survey such as the APMS (Crawford et al. Reference Crawford, Deary, Starr and Whalley2001; Bright et al. Reference Bright, Jaldow and Kopelman2002; Ali et al. Reference Ali, Ambler, Strydom, Rai, Cooper, McManus, Weich, Meltzer, Dein and Hassiotis2013). However, the NART scores are known to be less reliable at the extreme of the range, and it has been acknowledged that further work is required in assessing its validity in predicting IQ in individuals with borderline intellectual functioning (Nelson, Reference Nelson1991; Ali et al. Reference Ali, Ambler, Strydom, Rai, Cooper, McManus, Weich, Meltzer, Dein and Hassiotis2013). For the present study, we used the verbal IQ scores in both continuous and categorical forms. For our primary analysis, we standardized the verbal IQ variable to have a mean of 0 and a standard deviation of 1. For secondary analysis, we categorized the predicted WAIS-R IQ scores into three bands (>100, 86–100 and < 85) to offset any imprecision in scores at the extremes of the IQ distribution.
Sociodemographic characteristics
We included data on age (in years as a continuous variable), sex, highest education level [in three groups: (i) college or higher education, (ii) A-levels, O-levels, GCSEs or some qualifications and (iii) no qualifications]; accommodation tenure (owner occupier or living in rented accommodation); marital status [(i) married or cohabiting, (ii) single, widowed, divorced or separated] and occupational class based on the UK Registrar General's classification [coded as I/II (professionals or managerial professions), IIIa/IIIb (non-manual and manual skilled professions) and IV/V (semi-skilled or unskilled manual workers)].
Ethnicity
Participants were asked to identify their ethnic group on a showcard detailing 15 ethnic groups and an additional ‘other – please specify’ category. For the purpose of the current study, we coded the participants into two broad categories of White and non-White because of the relatively low numbers in the groups comprising the latter category.
Ascertainment of a common mental disorder
The revised Clinical Interview Schedule (CIS-R; Lewis et al. Reference Lewis, Pelosi, Araya and Dunn1992), a structured psychiatric interview designed to be used by trained non-clinical interviewers, was used to assess the presence of a common mental disorder. The recommended threshold of ⩾12 was used to identify people with enough neurotic symptoms to meet the ICD-10 diagnostic criteria (WHO, 1993) for one or more of the nine most common mental disorders including depressive, anxiety disorders and phobias (Lewis et al. Reference Lewis, Pelosi, Araya and Dunn1992).
Impulsivity
The Structured Clinical Interview for DSM-IV Axis II Disorders (SCID-II; Lobbestael et al. Reference Lobbestael, Leurgans and Arntz2011) was self-completed by all participants. We identified those describing themselves as impulsive by endorsing an item ‘Have you often done things impulsively’ in relation to being asked to report on the ‘kind of person you generally are’. Although this item seems to have good face validity, its properties in relation to other impulsivity screens is not known.
Ascertainment of hazardous and dependent alcohol use
Hazardous alcohol use was identified by the Alcohol Use Disorders Identification Test (AUDIT; Saunders et al. Reference Saunders, Aasland, Babor, de la Fuente and Grant1993) using the Computer-Assisted Self-completion Interview (CASI). The AUDIT includes questions about hazardous drinking in the past year (frequency and quantity of alcohol use), indicators of harmful alcohol use (feelings of guilt or remorse related to alcohol, blackouts, alcohol-induced injury, concerns about alcohol consumption) and symptoms of dependence (impaired control over drinking, salience of drinking, morning drinking). The Severity of Alcohol Dependence Questionnaire (SADQC; Stockwell et al. Reference Stockwell, Sitharthan, McGrath and Lang1994) was administered to all respondents scoring ⩾10 on the AUDIT and enabled identification of alcohol dependence in the past 6 months. To capture both hazardous drinking and alcohol dependence, we combined scores from the AUDIT and SADQC to produce a variable indicating (i) no hazardous alcohol use, (ii) hazardous alcohol use but no dependence and (iii) alcohol dependence.
Ascertainment of illicit drug use and smoking
Illicit drug use was also assessed using the CASI. Questions covered the past year use of eight drug types (cannabis, amphetamines, crack, cocaine, ecstasy, tranquilizers, opiates and volatile substances), followed by five questions based on the Diagnostic Interview Schedule (DIS; Malgady et al. Reference Malgady, Rogler and Tryon1992), designed to assess drug dependence. These included daily drug use for ⩾2 weeks, a sense of need or dependence, an inability to abstain, increasing tolerance and withdrawal symptoms. We coded respondents into (i) no drug use, (ii) any drug use and (iii) dependent drug use in the past year. Participants were asked a series of questions about the amount and frequency of tobacco smoking and we divided individuals into (i) non-smokers, (ii) occasional smokers and (iii) regular smokers (>7 cigarettes/week).
Analysis
All analyses were conducted using the ‘survey’ commands in Stata Statistical Software release 10.1 (Stata Corporation, USA). Probability weights derived by the APMS team were used to take account of the complex study design and non-response so that the results were representative in terms of age, sex, region and area characteristics of the household population aged ⩾16 years in England. First, sample weights were applied to take account of the different probabilities of selecting respondents in different sized households. Second, to reduce household non-response bias, a household-level weight was calculated from a logistic regression model using interviewer observation and area-level variables (based on Census 2001 data) available for responding and non-responding households. The non-response weight for each household was calculated as the inverse of the probability of response estimated from the model, multiplied by the household's selection weight. Third, weights were applied using calibration weighting based on age, sex and region to weight the data to represent the structure of the national population, taking account of differential non-response between regions, and age × sex groups. The population control totals used were the Office for National Statistics (ONS) 2006 mid-year household population estimates. Complete details of the weighting procedures are available elsewhere (McManus et al. Reference McManus, Meltzer, Brugha, Bebbington and Jenkins2009). We used the probability weights to estimate the 1-year prevalence of non-problem and problem gambling in the general population of England. In descriptive analyses, we cross-tabulated mean IQ scores and other putative correlates with gambling status (no gambling, non-problem gambling or problem gambling in the past year).
We used multinomial logistic regression models to estimate the relationship between each standard deviation decrease in IQ score and the odds of non-problem gambling and problem gambling, compared with no gambling in the past year. The other correlates identified in previous research, or a priori, were included in the models irrespective of their statistical significance. Similarly, we estimated crude and adjusted estimates of associations between these other potential correlates of problem gambling. We conducted our main analysis on a dataset with complete responses on IQ and gambling variables, and included covariates with missing data in the regression models by coding an additional ‘missing’ category.
We repeated the analysis as follows: (1) using the categorical measure of IQ, (2) repeating analyses on males only, (3) repeating analyses by restricting the sample to individuals aged < 65 years, (4) repeating all analyses on a dataset with complete data on all variables (i.e. dropping individuals with missing data in any of the covariates), and (5) repeating analyses after imputing missing values in each covariate with missing data with extreme values representing best- and worst-case scenarios for each covariate.
Ethical approval
Ethical approval for use of the APMS 2007 was obtained from the Royal Free Hospital and Medical School Research Ethics Committee (Ref: 06/Q0501/71). The current paper is based on publicly available, anonymized data for research use.
Results
Verbal IQ estimates were available for 6872 participants (92.8% of respondents). Complete data on gambling were available for 6827 of these individuals, who comprised our main analytic sample. Of this sample, 91.5% (n = 6291) had complete data on all covariates. More than two-thirds [68.0%, 95% confidence interval (CI) 66.7–69.2] of the sample (weighted to represent the general population of England) reported some form of gambling in the previous year. However, problem gambling (n = 36; weighted prevalence 0.7%, 95% CI 0.5–1.0) was rare, and pathological gambling rarer still (n = 15; weighted prevalence 0.3%, 95% CI 0.2–0.5).
Table 2 shows the characteristics of problem gamblers, non-problem gamblers and people reporting no gambling in the past year. Problem gamblers had a much lower estimated mean verbal IQ (87.5) than non-problem gamblers (mean IQ = 102.2) and non-gamblers (mean IQ = 103.3). Problem gamblers were also more likely than non-problem gamblers or non-gamblers to be younger males with a variety of socio-economic disadvantages, to be regular smokers, to endorse often making impulsive decisions, to be hazardous drug and alcohol users, and to have a common mental disorder (Table 2).
Descriptive statistics of respondents with problem gambling, non-problem gambling and no gambling in the past year
Values given as mean (standard deviation) or weighted percentages, rounded to one decimal place.
The total n (6827) represents respondents with complete data on IQ estimates and past year gambling.
Where missing data were present in a covariate, the percentage of these is reported within a separate row.
Table 3 shows the characteristics associated with problem and non-problem gambling (compared with no past-year gambling) estimated by multinomial logistic regression models adjusted for age and sex in model 1, and all the variables in model 2. Each standard deviation drop in estimated verbal IQ was associated with a twofold increase in the odds of problem gambling [adjusted odds ratio (OR) 2.1, 95% CI 1.3–3.4, p = 0.003]. Male sex, belonging to a social class other than I or II, White ethnicity, harmful use of alcohol, impulsivity and the presence of a common mental disorder were also associated with problem gambling in the adjusted analysis. By contrast, there was little evidence for any similar associations of non-problem gambling with lower IQ, and its associations with the other correlates were relatively weaker.
Multinomial logistic regression models to ascertain correlates of problem gambling and non-problem gambling compared with no past year gambling
s.d., Standard deviation.
Values given as odds ratios (95% confidence interval).
Model 1: adjusted for age and sex. Model 2: final model adjusted for all of the above variables.
* p < 0.05, ** p < 0.01, *** p < 0.001.
The associations between low verbal IQ and problem gambling remained robust in analysis using the categorical measure of IQ (Fig. 1). In the fully adjusted model, those with an estimated verbal IQ between 86 and 100 were four times (OR 4.1, 95% CI 1.4–12.5, p = 0.012) and those with an IQ < 85 were more than five times more likely (OR 5.6, 95% CI 1.9–16.7, p = 0.002) to be problem gamblers than those with an IQ of > 100, although the CIs were notably wide.
The odds of problem gambling by IQ in three categories using multinomial logistic regression.
The associations persisted when the analysis was repeated in males only, in those aged < 65 years, in a dataset with complete data on all covariates (n = 6291), and in sensitivity analyses with missing data in covariates imputed at either extreme representing the best- or worst-case scenario (Fig. 2).
The odds of problem gambling by each standard deviation decrease in IQ in sensitivity analyses using multinomial regression. The complete case dataset included only individuals without any missing data in any covariate (n = 6291). In the worst-case scenario individuals with missing data in: education were considered as having no education; tenure were coded as living in rented accommodation; occupational class were considered as being in class IV/V; ethnicity were considered as being non-White; impulsivity were considered as having endorsed impulsivity; smoking were considered as smokers of < 7 cigarettes/week; and illicit drug use were considered as dependent drug users. In the best-case scenario individuals with missing data in: education were considered as having a degree or higher education; tenure were coded as owner-occupiers; occupational class were considered as being in class I/II; ethnicity were considered as being White; impulsivity were considered as not having endorsed impulsivity; smoking were considered as non-smokers; and illicit drug use were considered as non-drug users. Worst- and best-case scenarios are arbitrary terms and do not reflect any value judgements. CI, Confidence interval.
Discussion
In this large population-based study in England, problem gambling was strongly associated with estimates of a lower verbal IQ, and the association persisted after controlling for sociodemographic characteristics and measures of mental health, impulsivity and a range of addictive behaviours. There was a twofold increase in the odds of problem gambling with each standard deviation drop in estimated IQ. This association was not present in the larger proportion of the population reporting any (non-problem) gambling in the past year and remained robust in sensitivity analyses.
To our knowledge, this is the first population-based study to have assessed the association between a measure of IQ and problem gambling. Such relationships have been reported as incidental findings in two small studies in clinical populations (Forbush et al. Reference Forbush, Shaw, Graeber, Hovick, Meyer, Moser, Bayless, Watson and Black2008; Kaare et al. Reference Kaare, Mottus and Konstabel2009) but these results were difficult to generalize because of the possibility of selection bias. Findings from an English cohort of 3750 young persons aged 17 years (the Avon Longitudinal Study of Parents and Children, ALSPAC) suggested that that an IQ of ⩽85 was significantly associated with participation in regular gambling (n = 354) (Emond et al. Reference Emond, Doerner and Griffiths2011). However, the numbers of individuals that could be defined as problem gamblers (n = 9) were too low for a meaningful comparison with IQ.
Although previous population-based studies have not investigated the role of IQ, there is a growing body of literature on the other correlates of problem gambling that we investigated. Our results are consistent with previously reported associations of problem or pathological gambling with a younger age, male sex, socio-economic adversities, common mental disorders, and alcohol and substance use disorders (Kessler et al. Reference Kessler, Hwang, LaBrie, Petukhova, Sampson, Winters and Shaffer2008; Brewer et al. Reference Brewer, Potenza and Desai2010; Lorains et al. Reference Lorains, Cowlishaw and Thomas2011). In the present study, White individuals were more likely to report any gambling in the past year, consistent with findings from the BGPS, although the findings of problem gambling being more common in non-White ethnic groups was not replicated (Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011). Longitudinal designs are required to clarify the direction of the reported associations because none of the above-mentioned cross-sectional studies, including ours, can make such inferences.
We found that almost two-thirds of the population reported gambling in the past year but less than 1% met the criteria for problem gambling. These estimates are consistent with those reported in the three previous BGPS that used identical criteria and a very similar methodology to our study (0.6% in 1999 and 2007 and 0.9% in 2010, with overlapping CIs) (Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011). These results are also similar to those from other European countries (Gotestam & Johansson, Reference Gotestam and Johansson2003; Bondolfi et al. Reference Bondolfi, Jermann, Ferrero, Zullino and Osiek2008; Bakken et al. Reference Bakken, Gotestam, Grawe, Wenzel and Oren2009), but lower than lifetime and past 12-month prevalence rates in the USA (Kessler et al. Reference Kessler, Hwang, LaBrie, Petukhova, Sampson, Winters and Shaffer2008) and Canada respectively (Cox et al. Reference Cox, Yu, Afifi and Ladouceur2005).
The main strengths of this study include the large population-based design, reducing the possibility of selection bias and ensuring high external validity of our results. Potential confounders including measures of mental health and addiction were measured using validated tools. The availability of such information is a significant advantage over the BGPS, which only collected data on demographic and socio-economic characteristics of problem gamblers (Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011). The use of probability weights to account for the sampling and non-response ensured that the sample, irrespective of non-response, was broadly representative of the general population of England. The CASI used in the gambling and addictions modules of the face-to-face interview (whereby the respondent inputs their answers directly onto a laptop computer rather than communicating them to the interviewer) is known to encourage honest answers to sensitive questions, and would have minimized the possibility of reporting bias (Tourangeau et al. Reference Tourangeau, Rips and Rasinski2000).
Our study has several limitations. First, despite the large sample, the low prevalence of pathological gambling (0.3%) in the survey meant that, as in the BGPS, we maximized statistical power by examining the correlates of problem gambling. Even at the relaxed threshold, the low numbers resulted in relatively wide CIs, a problem common to almost all other population-based studies of problem gambling (Wardle et al. Reference Wardle, Moody, Spence, Orford, Volberg, Jotangia, Griffiths, Hussey and Dobbie2011). Second, although the problem gambling screen based upon DSM-IV criteria has been used in previous large-scale surveys, it is not clear that all who screen positive will require clinical intervention. The screen assumes that problem gambling occurs on a continuum of severity with the more severe and clinically defined category of pathological gambling. Furthermore, it is possible that individuals with lower cognitive abilities interpret and answer questions related to problematic gambling differently from those with higher intellectual abilities, although it should be noted that individuals with lower IQs were no more likely to report non-problem gambling than those with higher IQ. Third, although selective non-participation of people with gambling problems is possible, it may be less likely in the APMS (where the gambling screen was a small part of a larger mental health survey) than in special purpose gambling surveys. Fourth, the NART, although highly correlated with verbal and full-scale IQ (Bright et al. Reference Bright, Jaldow and Kopelman2002), cannot give information on any functional limitations associated with impaired cognitive abilities. Importantly, the NART itself has not been validated for individuals with intellectual disabilities; and the private household survey design and exclusion criteria for the NART may have precluded individuals with significant levels of intellectual disabilities from participating. Finally, although we had access to a large number of potential confounders including impulsivity and addictive behaviours, the possibility of residual confounding cannot be excluded. Such confounding may arise from inadequate measurement of variables such as impulsivity, which in the current study was assessed using a single item, or from other unmeasured confounders, including characteristics related to early life and development.
Implications and future directions
We found a lower estimated verbal IQ to be associated with problem gambling even after accounting for several potentially confounding sociodemographic characteristics, mental health, impulsivity and other addictive behaviours. The lack of similar associations in people with non-problem gambling may point towards general cognitive abilities being a predictor of individuals at a higher risk of making a transition to problem gambling. The results warrant future research being carried out to replicate these findings, to understand the underlying mechanisms. Our findings also highlight that efforts should be made to validate gambling screens in individuals with lower intellectual abilities.
Neuropsychological studies have previously reported that problem or pathological gamblers are more likely to engage in impulsive behaviours (Goudriaan et al. Reference Goudriaan, Oosterlaan, deBeurs and van den Brink2004; Verdejo-Garcia et al. Reference Verdejo-Garcia, Lawrence and Clark2008), have impaired decision making (Goudriaan et al. Reference Goudriaan, Oosterlaan, deBeurs and van den Brink2004, Reference Goudriaan, Oosterlaan, deBeurs and van den Brink2005), steeper associations with delay discounting (Dixon et al. Reference Dixon, Marley and Jacobs2003; Goudriaan et al. Reference Goudriaan, Oosterlaan, deBeurs and van den Brink2004) and other deficits in cognitive reward pathways including a failure to learn from mistakes (Goudriaan et al. Reference Goudriaan, Oosterlaan, deBeurs and van den Brink2004). These are obvious avenues to explore in future research, ensuring cognitive abilities of individuals are accounted for when assessing potential pathways towards problem or pathological gambling.
It is known that problem gamblers rarely seek help or treatment (Bowden-Jones & George, Reference Bowden-Jones and George2011; Shaffer & Martin, Reference Shaffer and Martin2011), but in our experience gambling problems are not usually considered in clinical practice, even when alcohol and drug misuse are the presenting symptoms. Our findings suggest that clinicians working with individuals with lower intellectual functioning should be mindful of this potential hidden vulnerability in these individuals. Conversely, IQ tests should be considered in the work-up of problem gamblers and may facilitate the appropriate and accessible delivery of interventions and enhance engagement.
Finally, gambling opportunities are widespread in many Western societies, but relatively little policy attention has been paid to preventing gambling problems. Warnings of problems associated with gambling are typically ‘small print’, and gambling opportunities are widely advertised and promoted in print, television, radio and the internet. Our findings, if confirmed, raise important questions about the impact of such coverage and the lack of more accessible warnings for vulnerable members of society, particularly those of more limited intellectual ability.
Acknowledgements
No specific funding was received for the current study. The APMS 2007 was carried out by the NatCen in collaboration with the University of Leicester, and was commissioned by the National Health Service (NHS) Information Centre for Health and Social Care with funds from the Department of Health. W. Hall is funded by a National Health and Medical Research Council Australia Fellowship (grant ID: 569 738). P. Moran is part-funded by the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and the Maudsley NHS Foundation Trust and King's College London.
We sadly report that Howard Meltzer, our co-author, colleague and friend, died on 23 January 2013.
Declaration of Interest
None.
