Adolescent Problem Gambling in Rural Ghana: Prevalence and Gender Differentiation

We conducted a cross-sectional survey, using a self-report questionnaire, among regularly attending Junior High Schools³ (JHS) adolescents in the Ayensuano rural district in the eastern region of Ghana. A cross-sectional survey is appropriate for the retrospective estimation of the prevalence an outcome variable, and for the identification of the correlates of an outcome (Woodward 2014). We obtained the list of schools and pupil statistics from the District Educational Directorate of Ayensuano: there were 84 JHS being attended by 9365 students. Based on an estimated prevalence of 42% of mobile gambling among young people in Ghana (GeoPoll 2017), we predetermined a minimum sample size of 971 students, using Cochran’s formula for calculating a sample for proportions (Cochran 1963). However, the final predetermined sample (n = 1214) included an additional 25% for two reasons: to obtain satisfactory precision and confidence interval widths, and to compensate for nonresponse (Naing et al. 2006). In all, we randomly selected and approached 15 schools proportional to enrollment size to participate in the study; however, the heads of 12 schools permitted the study in their schools (representing a school response of 80%). Across the 12 selected participating schools, we randomly selected 36 classes proportional to enrollment size. In each selected class, all the students qualified to respond to the survey–we invited all the students in a selected class to participate in the survey. Students who were willing to respond to the survey and provided their consent were included, whereas students who were absent on the day of the survey were excluded from the study. In all, we approached and invited 1214 students to respond to the survey, but, 1101 adolescents (female = 575; male = 526) provided complete data included in the study–representing a response of rate of 90%.

We included seven items to assess the social characteristics and demographic background of the participants: gender (female or male), age, grade, living arrangement, caretaker’s employment status, family structure (measured by father’s number of wives), and romantic relationship status (see Table 1).

We used the Diagnostic and Statistical Manual of Mental Health Fourth Edition Adapted for Juveniles (DSM-IV-MR-J) questionnaire (Fisher 2000) to assess problem gambling during the previous 12 months. The DSM-IV-J is a 12 item checklist which assesses nine criteria of problem gambling: progression and preoccupation, tolerance, withdrawal and loss of control, escape, chasing, lies and deception, illegal activities, family and school disruptions, and financial bailout (Jacques and Ladouceur 2003). Predominantly, the items are on a 4 point rated scale: never, once or twice, sometimes, and often. The presence of four (or more) out of the nine criteria indicates problem gambling (Fisher 2000; Jacques and Ladouceur 2003). Example, In the past year have you ever spent much more than you planned to on gambling? Answers are transformed into dichotomous responses (presence–absence of problem gambling, based on the criteria).The DSM-IV-J scale has an acceptable internal consistency (Cronbach’s alpha = 0.75) (Fisher 2000) and has been used extensively across the world (Derevensky and Gupta 2005b; Gupta and Derevensky 2000). In sub-Saharan Africa, the scale has demonstrated satisfactory validity and reliability (Abdi et al. 2015; Ssewanyana et al. 2018). The Cronbach’s alpha score in the current study was 0.81.

We included eight binary (No or Yes) response rated personal lifestyle factors and variables assessing the experience of interpersonal adversities, adopted from the 2012 WHO–Global School-based Student Health Survey questionnaire used in Ghana (Owusu 2012). These variables included: weekly alcohol use (In a typical week, how many times do you have at least on alcoholic drink?), truancy (During the past 12 months, how many days did you miss classes or school without permission?), bullying victimisation (During the past 12 months, how many days were you bullied?), and sexual abuse victimisation (Has anyone forced you [i.e. physically or verbally] to engage in sexual activities against your will?). We also adopted one item from the 5 item Duke University Religion Index (DUREL) (Koenig and Büssing 2010) to assess religious participation: “How often do you attend church or other religious meetings?” with response rating options ranging from (1) ‘never’ to (6) ‘more than once per week’. We have shown in the supplementary material the variables included in this study and the specific questions used to assess them in the survey (eTable 1).

The Statistical Package for Social Sciences (SPSS version 26.0 for Windows) was used for the analysis. Gambling (non-problem gambling vs. problem gambling) was the outcome variable, while the exposure variables or correlates were composed of the socio-demographic variables, personal factors and the variables of social adversity specified. As the loss of cases due to missing data was less than 5%, we used the list‐wise deletion of missing data strategy (Graham 2009). We have shown in the supplementary material the coding of the variables included in our statistical analysis (eTable 1). The data analysis proceeded in three stages; at each stage, we stratified the data by gender (female and male), guided by the aims of the study. Firstly, we performed descriptive analysis of the data by applying frequencies, proportions and the frequentist 95% confidence intervals (CI) (Hespanhol et al. 2019) to assess the uncertainty around the effect estimates of the difference in proportions (DP) between female and male adolescents in terms of social demographic variables and the 12-month prevalence estimates of problem gambling. In stage two, given the categorical nature of the data, we applied the Pearson’s Chi-squared test (\({\upchi })\) to explore the bivariate relationships between problem gambling and each of the socio-demographic variables, personal factors and the variables of social adversity included in the study. We performed a point-biserial correlation (r_pb) test to examine the possible bivariate relationship between religious participation and problem gambling (Prematunga 2012). Statistically significant results were determined using the p value less than 0.05 (p < 0.05).The final stage of the analysis involved multivariable logistic regression, to examine the possible associations between the binary outcome variable (gambling: non-problem gambling vs. problem gambling) and the specified correlates (socio-demographic variables, personal factors, and variables of social adversity). We built three models, one each for the overall sample, female adolescent sub-sample, and the male adolescent sub-sample. As recommended by leading logistic regression modelling experts, the candidate correlates were entered in the multivariable logistic regression models regardless of the statistical significance of their bivariate relationship with the outcome variable (Babyak 2004). We reported the results of the logistic regression as odds ratios with 95% confidence intervals (CI) and p values (Greenland et al. 2016).

The 1101 participants in this study were aged 10–19 years (mean = 15.3; modal = 16; SD = 1.9). There were more female (n = 575) than male (n = 526) adolescents. Table 1 presents the socio-demographic characteristics of the participants, stratified by gender.

While the majority of younger adolescents were females (n = 294 [DP 0.089, 95% CI 0.03, 0.15]), most of the older adolescents were males (n = 304 [DP − 0.089, 95% CI − 0.15, − 0.03]). The majority of the participants were in JHS 1 (64.3%), lived with both parents (55%), and identified their family as monogamous (63.8%). More females (n = 270) than males (n = 199) indicated as being in a romantic relationship (DP 0.36, 95% CI 0.03, 0.15).

Overall, 378 (34.3%) participants reported problem gambling during the previous 12 months. Regarding gender, more male adolescents (n = 204; 38.8%) than females (n = 174; 30.3%) reported problem gambling during the previous 12 months [DP 0.08, 95% CI 0.029, 0.141]. In terms of age, more younger adolescents (n = 200; 38.8%) reported problem gambling during the previous 12 months, compared to older adolescents (n = 178; 30.4%) [DP 0.08, 95% CI 0.027, 0.14]. The prevalence estimates of problem gambling were varied across grade: JHS 1 (n = 242; 34.2%), JHS 2 (n = 108; 38.7%), and JHS 3 (n = 28; 24.6%).

Across the overall sample and gender, we found statistically significant bivariate associations between problem gambling and most of the socio-demographic variables, personal factors and social adversities included in this study (see Table 2).

Males were more likely to report problem gambling than female adolescents (\({\upchi }^{2}\)₍₁₎ = 8.85, p = 0.003). Comparatively, four variables showed the strongest statistical bivariate correlation with problem gambling across the total sample and gender: sexual abuse victimisation (\({\upchi }^{2}\)₍₁₎ = 119.0, p < 0.001), breakup (\({\upchi }^{2}\)₍₁₎ = 73.4, p < 0.001), weekly alcohol use (\({\upchi }^{2}\)₍₁₎ = 72.0, p < 0.001), and being in a romantic relationship (\({\upchi }^{2}\)₍₁₎ = 61.2, p < 0.001).

Furthermore, across the overall sample, there was a statistically significant negative correlation between religious participation and problem gambling during the previous 12 months (r_pb = − 0.74, N = 1090, p = 0.015); however, when stratified by gender, there was a statistically non-significant negative correlation between religious participation and problem gambling neither among females (r_pb =  − 0.080, N = 571, p = 0.057) nor males (r_pb = − 0.054, N = 519, p = 0.217).

Each of the final logistic regression models for the overall sample (\({\upchi }\) _{(df = 18)} = 272.54, p < 0.001), female sub-sample (\({\upchi }\) _{(df = 17)} = 146.84, p < 0.001), and the male sub-sample (\({\upchi }\) _{(df = 17)} = 130.22, p < 0.001) was statistically significant, accounting for 75%, 76%, and 74% of the variance within the outcome variable in the overall, female, and male models respectively. Table 3 shows the results of the multivariable analysis of demographic and exposure variables associated with problem gambling in the previous 12 months, stratified by gender.

As shown in Table 3, sexual abuse victimisation (AOR = 2.99; 95% CI = 2.17, 4.12), weekly alcohol use (AOR = 1.94; 95% CI = 1.37, 2.73), male gender (AOR = 1.89; 95% CI = 1.40, 2.56), breakup (AOR = 1.63; 95% CI = 1.17, 2.25), bullying victimisation (AOR = 1.59; 95% CI = 1.18, 2.15), being in a romantic relationship (AOR = 1.51; 95% CI = 1.09, 2.09), parental divorce (AOR = 1.43; 95% CI = 1.02, 2.00), and being in a polygamous family—father having more one wife (AOR = 1.41; 95% CI = 1.03, 1.93) were significantly associated with increased odds of problem gambling. However, religious participation (AOR = 0.90; 95% CI = 0.82, 0.99), being an older adolescent (AOR = 0.69; 95% CI = 0.51, 0.93) and being in JHS 3 (AOR = 0.49; 95% CI = 0.28, 0.84) were associated with reduced odds of problem gambling.

Among females, sexual abuse victimisation (AOR = 3.43; 95% CI = 2.15, 5.48), breakup (AOR = 2.01; 95% CI = 1.24, 3.26), parental divorce (AOR = 2.00; 95% CI = 1.22, 3.26), and being in a polygamous family (AOR = 1.64; 95% CI = 1.02, 2.63) were associated with increased odds of problem gambling; however, being in JHS 3 (AOR = 0.43; 95% CI = 0.19, 0.96) was associated with reduced odds of problem gambling–see Table 3.

Table 3 also shows that among male adolescents, whereas sexual abuse victimisation (AOR = 2.81; 95% CI = 1.75, 4.53), weekly alcohol use (AOR = 2.23; 95% CI = 1.34, 3.70), being in a romantic relationship (AOR = 1.99; 95% CI = 1.25, 3.15), and bullying victimisation (AOR = 1.74; 95% CI = 1.12, 2.68), were associated with increased odds of problem gambling, being an older adolescent (AOR = 0.61; 95% CI = 0.40, 0.93) was associated with reduced odds of problem gambling.

Thus, it is notable that sexual abuse victimisation emerged as the strongest and only factor associated with the increased odds of problem gambling across the overall sample (AOR = 2.99; 95% CI = 2.17, 4.12), among females (AOR = 3.43; 95% CI = 2.15, 5.48) and male adolescents (AOR = 2.81; 95% CI = 1.75, 4.53).

Generally, the 12 month prevalence estimate of problem gambling found in this study (34.3%) is higher, relative to estimates from high-income countries, which range from 0.2% to 12.3% (Calado et al. 2017; Hardoon et al. 2004). However, as should be expected, the reported prevalence estimates of the current study are comparable to estimates from other countries within sub-Saharan Africa, including Nigeria (Aguocha et al. 2019) and Ethiopia (Abdi et al. 2015). Again, consistent with other studies from the subregion (Abdi et al. 2015; Aguocha et al. 2019; Glozah et al. 2019; Sharp et al. 2015). Beyond possible individual-level motivational factors, the higher prevalence estimates in current study could be attributable to the social acceptability of gambling for the purposes of financial gains and the relatively increased adolescent access to gambling centres or sources in rural Ghana (Department of Social Welfare 2018; Hayk and Sailer 2020). The Gaming Act of Ghana⁴ prohibits children from gambling and forbids operators of gambling centres and machines from allowing children to patronise. However, a recent report by the Department of Social Welfare in the Ayensuano rural district (where this study was conducted) suggests that child and adolescent gambling is now a public concern and thus recommends that, “Game center operators should be severely dealt with since adolescents patronize it… Chinese Jackpot games should be seized” by the authorities (p.6) (Department of Social Welfare 2018). Also, the most recent published evidence from rural Ghana suggests that Chinese slot machines are a bane of the life of rural-dwelling young people (Hayk and Sailer 2020). Given this contextual reality, we are not entirely surprised at the prevalence estimate of probable problem gambling in this study, even though it appears comparatively higher.

Generally, the correlates of problem gambling found in this study (sexual abuse victimisation, weekly alcohol use, male gender, breakup, bullying victimisation, being in a romantic relationship, parental divorce and being in a polygamous family) are consistent with evidence in the global literature (Derevensky et al. 2011; Floros 2018; Grande-Gosende et al. 2020; Hayer and Griffiths 2015; Lane et al. 2016; Molinaro et al. 2014; Shead et al. 2011). Specifically, this study shows that regardless of gender, participants who reported sexual abuse victimisation were about three times more likely to experience problem gambling. This finding is consistent with an important evidence in the area, that adolescent victims of sexual abuse are at elevated risk of problem gambling. It has been argued that adolescent victims of sexual abuse are more likely to take to gambling as a means of emotional coping, given that the problem solving capacities and coping skills are not fully formed during adolescence (Blaszczynski and Nower 2002; Felsher et al. 2010; Lane et al. 2016). Similarly, female adolescents who report a breakup and parental divorce are likely to gamble as a means of emotional coping and ‘escaping’ their psychological and social pains. In other words, adolescent problem gambling can be interpreted as a response to traumatic experiences (Derevensky et al. 2011; Felsher et al. 2010; Messerlian et al. 2005). A notable finding of this study is that females from polygamous families are more likely to report problem gambling. This evidence is to be expected, as young girls in polygamous families tend to be the most disadvantaged; there is evidence to suggest that compared to boys, girls in polygamous families experience more corporal punishment and lower academic performance (Al-Sharfi et al. 2016). Even so, considering that most polygamous families are poor, it is not readily clear from this study whether adolescent girls from polygamous families engage in gambling with the motive to make money to support themselves or as a means of coping or escaping the social adversities in larger families. Perhaps qualitative studies may prove more informative in exploring the firsthand accounts and individualised meanings of female adolescents from polygamous families.

The findings of this study show weekly alcohol use as a strong positive correlate of problem gambling among male adolescents. Whereas this evidence is consistent with reports from both high-income and low- and middle-income contexts (Aguocha et al. 2019; Shead et al. 2010), the explanation of the exact risk and causal relationships between alcohol use and gambling is not straightforward, even though previous evidence has linked alcohol use to impulsivity and health risk-taking behaviours among adolescents (Inchley et al. 2018). Problem gambling could be related to the odds of alcohol use, as problem gamblers tend to experience problem solving and emotion regulation difficulties (Jauregui et al. 2016), and alcohol use could be a factor for problem gambling, but each of these factors could also be outcomes influenced separately by other factors (Barnes et al. 1999; Cronce and Corbin 2011; Huggett et al. 2019; Jauregui et al. 2016; Sagoe et al. 2017). A recent twin study suggests that shared genetic and (or) risk factors in the environment influence the association between frequent alcohol use and increased gambling (Huggett et al. 2019). Furthermore, being in a romantic relationship was associated with increased odds of problem gambling among male adolescents. Typically, in Ghana, adolescent males in heterosexual romantic relationships assume the role of providing financial support to their girlfriends; girlfriends are financially dependent on their boyfriends (Ampofo 2001). This implies that, perhaps, for an unemployed school-going male adolescent, gambling could represent a source of earning some money to provide financial support to their girlfriend. Future studies could explore further the key motivational factors undergirding problem gambling among in-school male adolescents who are in romantic relationships.

Three factors, even though with relatively lower effect sizes, showed associations with reduced odds of problem gambling: religious participation/attendance, older age, and being in a higher school grade. Older adolescents in Junior High School grade 3, who participated in religious activities or attended religious services frequently were less likely to report problem gambling. Whereas this evidence supports previous findings in the area (Dowling et al. 2017; Floros 2018), our results do not demonstrate whether these behavioral factors are key ‘protective’ variables against the onset of problem gambling.

This study is partly in response to the call for expansive studies on problem gambling among young people in countries within sub-Saharan Africa (Ssewanyana and Bitanihirwe 2018). This study represents the first attempt from a sub-Saharan Africa country at providing a systematic evidence on the prevalence of problem gambling and some personal and interpersonal correlates among rural in-school adolescents. Nonetheless, some plausible limitations of the study are noteworthy. The multiple responses of the DSM-IV-MR-J scale are collapsed when computing the scores; this could lead to insufficient evidence in support of the classification accuracy of the scale (Derevensky and Gupta 2005b; Edgren et al. 2016; Stinchfield 2011). Relatedly, as adolescent gambling represents a relatively new area of research (Blinn-Pike et al. 2010), most of the current screening measures (including DSM-IV-MR-J scale) of gambling in young people lack sufficient construct validity (Blinn-Pike et al., 2010; Derevensky and Gupta 2000, 2005b). It has been suggested, for example, that the DSM-IV-MR-J scale results in overestimates of the prevalence of probable problem gambling among young people (Derevensky et al. 2003). However, while we acknowledge that our reported prevalence estimates may be inflated, we also agree with the argument by leading youth gambling researchers that when the number of adolescents experiencing gambling challenges is as high as reported in research, more adolescents are likely to present for treatment (Derevensky et al. 2003).

Furthermore, there is evidence to suggest that studies on gambling among young people are susceptible to social desirability bias (i.e., self-deceptive enhancement effects, and impression management effects) (Kuentzel et al. 2008). However, in our study we believe that this bias might be low, as participants were allowed enough privacy in responding to the survey: we used an anonymous self-report questionnaire, participants sat as far apart from one another as possible while answering the questionnaire, and teachers were kept in the background during the survey. It is also notable that causal interpretation of our findings is not possible, given the cross-sectional design used (Woodward 2014). Lastly, our study failed to include absentee students on the day of the study and out-of-school rural-dwelling adolescents. This implies that whereas our findings may be generalisable across non-clinical sample of in-school adolescents in rural eastern Ghana (and other rural contexts in sub-Saharan Africa), they may not necessarily apply to their peers who are out of school. Besides not including an urban adolescent sample to facilitate a comparative ecological analysis in this study, our study also failed to consider some variables which could present as covariates of probable problem gambling among adolescents. For example, personality factors, anxiety, depression, social acceptability, family/parental permissiveness of adolescent gambling, access and proximity to gambling points and sources, among personal, family and locality factors (Derevensky et al. 2011; Derevensky and Gupta 2005a).

Clearly, more research is needed to expand the evidence base on adolescent (problem) gambling in rural Ghana, as the findings of this study come from only one rural district. Specifically, cross-sectional surveys and interview-based studies that identify the motivations of rural adolescents for gambling are needed, but also longitudinal studies using robust designs are needed to understand the risk and protective factors, and patterns and changes in gambling behaviour across adolescence through early adulthood. Potentially, future studies of this nature would be more useful in informing intervention, prevention and health promotion programmes. As recommended by recent systematic reviews (Floros 2018; Keen et al. 2017; Ladouceur et al. 2013), the position of this study is that any intervention and prevention efforts and treatment models should be theory-driven and evidence-based. Drawing on the evidence of this study, school-based psychoeducation and harm reduction prevention programmes could be instituted in schools within rural Ghana, following universal, selective, and indicative prevention frameworks. This approach has been found to be potentially useful in the treatment and prevention of gambling among young people (Dickson et al. 2004; Floros 2018; Ladouceur et al. 2003). The evidence in this study also underscores the need for the government of Ghana (through the Ghana Police Service and the Department of Social Welfare) to enforce strictly the abstinence law on underage gambling, but also pursue and implement the relevant educational and social policies aimed at enhancing supportive school climate, mitigating familial adversities, and protecting adolescents (particularly, young girls) from abuses.