II. The utility of outcome expectancies in the prediction of adolescent gambling behaviour

Meredith A. M. Gillespie, Jeffrey Derevensky, & Rina Gupta, International Centre for Youth Gambling Problems and High-Risk Behaviors, Montreal, Quebec, Canada.
E-mail: merdith.gillespie@mail.mcgill.ca

Abstract

The Gambling Expectancy Questionnaire (GEQ; Gillespie, Derevensky & Gupta, 2006, previous article) suggests that adolescents hold a variety of positive and negative outcome expectancies related to gambling. Significant age, gender, and DSM-IV-MR-J gambling group differences were identified on the scales of the GEQ (i.e., enjoyment/arousal, self-enhancement, money, overinvolvement, emotional impact) in this study. Direct logistic regression among adolescent gamblers was performed separately for males and females to predict group membership in either social or problem gambling categories. The results provide insightful information suggesting that non-gamblers, social gamblers, at-risk gamblers, and probable pathological gamblers (PPGs) differ in the strength of their expectancies of both the positive and negative outcomes of gambling behaviour. In particular, PPGs highly anticipate both the positive and negative outcomes of gambling. Among males, these perceptions differentiate those who gamble excessively and those who do not. For females, outcome expectancies may have less predictive value. These findings were interpreted in terms of their implications for prevention, treatment, and future research.
Key words: youth gambling, outcome expectancy, perceived benefits and risks

Introduction

Social cognitive models of health behaviour (e.g., Health Belief Model, Becker, 1974; Theory of Planned Behavior, Ajzen, 1991) place importance on the subjective cognitions implicated in behaviour choice. Some researchers have argued that youth engage in potentially risky behaviours, like gambling, primarily because of the perceived benefits (e.g., pleasure, entertainment, excitement, peer approval, and relaxation) (Moore & Gullone, 1996). Accordingly, adolescents may fail to consider the potential costs and negative consequences of such behaviour, thereby underestimating the related risks (Clayton, 1992). Thus, in keeping with social cognition theories, an individual's decision to engage in gambling likely reflects the differential salience of its positive and negative outcomes. What youth expect to gain (i.e., positive expectancies) as well as what they expect to lose (i.e., negative expectancies) from their gambling is likely to play a significant role in their decisions to initiate and maintain their gambling behaviour.

Recent studies of drug and alcohol outcome expectancies suggest that the beliefs and perceptions an adolescent holds about the positive and negative outcomes of drugs or alcohol use play a critical role in their decisions to initiate and to maintain these high-risk behaviours (Brown, Christiansen, & Goldman, 1987; Fromme & D'Amico, 2000; Goldberg & Fischhoff, 2000; Goldberg, Halpern-Felsher, & Millstein, 2002; Johnston, 2003; Johnston, O'Malley, & Bachman, 2001; Leigh & Stacy, 1993; Stacy, Widaman, & Marlatt, 1990). In particular, outcome expectancies have been shown to play an integral role in the maintenance of alcohol use, and they have been used to predict how serious an individual's involvement in a high-risk activity may become (Brown et al., 1987; Fromme & D'Amico, 2000; Goldberg & Fischhoff, 2000; Leigh & Stacy, 1993; Stacy et al., 1990). More specifically, much of the adolescent alcohol literature highlights positive expectancies (i.e., beliefs about the beneficial effects of alcohol) as better predictors of teen alcohol consumption than negative expectancies (Goldberg et al., 2002; Stacy et al., 1990). The more positive one's expectations of the outcome of drinking behaviour, the more heavily one drinks, and the greater the likelihood for alcohol-related problems (Fromme & D'Amico, 2000).

To date, little research has explored adolescents' perceptions of the consequences of gambling behaviour. Likewise, very few studies have directly examined how these positive and negative outcome expectancies influence adolescent gambling participation. Although the identification of gambling outcome expectancies is only one small piece of the much larger puzzle of predicting and preventing problem gambling, it is a piece that is currently missing. As such, its exploration as a line of inquiry may have the potential to inform future prevention and treatment initiatives.

As a means to extend outcome expectancy research into the field of youth gambling, Gillespie, Derevensky, and Gupta (2006, previous article) recently sought to develop a Gambling Expectancy Questionnaire (GEQ) that could evaluate the strength of adolescents' positive and negative outcome expectancies of gambling. Alcohol expectancy instruments served as a template for the development of the instrument. From an analysis of adolescents' endorsements of 48 gambling expectancy items, representing the diversity of gambling's biopsychosocial risks and benefits (American Psychiatric Association, 1994; Fisher, 2000; Griffiths & Delfabbro, 2001; Gupta & Derevensky, 1998a; Neighbours, Lostutter, Cronce, & Larimer, 2002), five distinct outcome expectancy constructs emerged and thus were represented as the five scales of the GEQ. Adolescents perceived enjoyment/arousal, self-enhancement, and money as salient yet discrete positive outcomes of gambling. In other words, youth anticipate a combination of enjoyment, excitement, and social opportunities from gambling (i.e., enjoyment/arousal). They also perceive gambling as an opportunity to feel good about themselves, either by impressing their peers or by establishing autonomy from others (i.e., self-enhancement). Moreover, they anticipate making money from gambling activities (i.e., money). Conversely, adolescents also perceived two distinct negative outcomes associated with gambling. Adolescents' responses reflected their understanding of the potential for preoccupation with gambling and the relational disruptions that may take place as a consequence (i.e., overinvolvement). They also clearly anticipated a potential negative emotional impact from gambling (i.e., emotional impact).

The recent development of the GEQ provides an opportunity to explore the salience of these outcome expectancies for adolescents differing in age, gender, and gambling severity. While the predictive utility of expectancy models has been well documented in relation to alcohol and drug use, both from an applied and a preventative research perspective, virtually no studies have empirically examined how outcome expectancies operate to predict gambling severity among adolescents. Given the commonalities found in the risk and protective factors among adolescent alcohol use, drug use, and gambling behaviour (Dickson, Derevensky, & Gupta, 2002), it is reasonable to suggest that the positive and negative effects that adolescents associate with gambling may help predict excessive gambling behaviour. It is expected that youth gambling outcome expectancies will differ among those who gamble excessively, those who are able to gamble responsibly, and those who choose not to gamble at all. Similarly, these behaviour-specific cognitions may differentiate social gamblers (i.e., non-problem gamblers) and problem gamblers.

Method

Participants

Participants were 1,013 students (males = 432 (42.6%); females = 581 (57.4%)) from grades 7 to 11 (age range = 11–18; mean age = 14.77 years; SD = 1.52). The majority of these students resided in the greater Montreal area, with approximately 6% of the sample being obtained in the Ottawa area. The majority (99.1%) of the sample was 17 years of age or younger, and thus legally prohibited from gambling on provincially regulated forms of gambling. Only 0.9% of the sample was of legal age to participate in provincially regulated gambling activities.

Approval was requested and obtained from four school boards in the greater Montreal area for participation. Individual high schools were then approached with a detailed proposal once school board approval was granted. In total, nine public high schools approved their students' participation in the study. Students from three private schools in Montreal and one private school in Ottawa were also included. A total of 13 schools, located in both urban and suburban areas and representing considerable variability in socioeconomic and cultural backgrounds, were included in this study.

Measures

Gambling Activities Questionnaire—Adapted (GAQ) (Gupta & Derevensky, 1996). The GAQ is designed to assess four general domains related to gambling behaviour: descriptive information including prevalence, types of activities, frequency of gambling, amount wagered, and social factors; cognitive perceptions about the amount of skill and luck involved in various gambling and nongambling activities; familial gambling and parental gambling behaviour; and comorbidity with other addictive and delinquent behaviours. For this study, a modified version of the GAQ was employed that included descriptive information regarding the frequency of gambling behaviour across various types of activities.

DSM-IV-MR-J (Fisher, 2000). This 12-item, 9-category instrument is a screen for pathological gambling during adolescence. It has been modeled upon the DSM-IV (APA, 1994) criteria for diagnosis of adult pathological gambling. An earlier version (DSM-IV-J) (Fisher, 1992) has been used by several researchers and was found to be the most conservative measure of pathological gambling among adolescents (Derevensky & Gupta, 2000; Gupta & Derevensky, 1998a, 1998b; Marget, Gupta, & Derevensky, 1999; Powell, Hardoon, Derevensky, & Gupta, 1999; Volberg, 1998). The revised version, DSM-IV-MR-J (MR = multiple response, J = juvenile) was developed for use with adolescents that have gambled over the past year. It assesses a number of important variables related to pathological gambling: progression, preoccupation, tolerance, withdrawal, loss of control, escape, chasing losses, deception, illegal activity, and family/school disruption.

GEQ (Gillespie et al., 2006). The 23-item GEQ comprises five discrete scales representing three positive outcome expectancies—enjoyment/arousal (a = .86), self-enhancement (a = .81), and money (a = .78)—and two negative outcome expectancies—overinvolvement (a = .91) and emotional impact (a = .85). For each scale, items are scored on a 7-point Likert scale ranging from 1 (no chance) to 7 (certain to happen), with a neutral middle point 4 (neither likely nor unlikely). The enjoyment/arousal scale consists of eight items denoting enjoyment, excitement/arousal, boredom, escape/tension reduction, and social interaction. The self-enhancement scale includes four items representing the themes of social acceptance and independence, while the money scale consists of three items denoting the theme of gambling to make money. The overinvolvement scale is composed of five items representing the negative themes of preoccupation and relational disruptions and the emotional impact scale is composed of three items denoting gambling's negative emotional effects. As a result of the combination of benefit and risk themes comprising each of its five subscales, the GEQ reflects the intricacy of adolescents' gambling outcome expectancies.

Procedure

The GEQ was group-administered to participants in classrooms and/or conference rooms by several trained research assistants. Groups ranged from 10 to 60 students, with the number of research assistants varying according to group size. Students were provided with a brief description of the types of questions that would be asked (e.g., "Some questions will ask you about your gambling behaviour; some questions will ask you about what you expect to happen when you gamble") as well as instructions regarding the completion of the instrument ("Please make sure to take your time and read all the questions and instructions carefully. Also make sure to fill in the circles completely with the pencil that has been provided"). Students were also given the following definition of gambling to keep in mind when they responded: "Gambling is any activity that you play in which you are putting money, or something of monetary value, at risk since winning and/or losing is based on chance."

Results

Data analyses

The prevalence of gambling participation among adolescents was analyzed using descriptive statistics. For these analyses, the age variable was recoded into two categories: younger adolescents (11–14 years; n = 391) and older adolescents (15–18 years; n = 617). A 2 (gender) × 4 (DSM groups) × 2 (age) factorial analysis of variance was performed in order to assess group differences on the five scales of the GEQ: enjoyment/arousal, self-enhancement, money, overinvolvement, and emotional impact. The Dunnett's C Post Hoc test, which does not assume equality of variances, was used to compare mean differences between students based upon four gambling categories: non-gamblers, social gamblers (DSM-IV-MR-J = 0–1), at-risk gamblers (DSM-IV-MR-J = 2–3), and probable pathological gamblers (PPGs) (DSM-IV-MR-J ≥ 4). Since one factorial ANOVA was performed for each scale (total = 5), the alpha level was set at p < .01 for each analysis. Nonparametric tests were used to validate the findings of the univariate analyses due to the nonnormal distributions of the five GEQ scales. The Kruskal–Wallis statistic was used to test differences based on the severity of gambling problems, and a two-sample Kolmorov–Smirnov test was used for gender and age variables. All of the nonparametric tests yielded the same results as the parametric tests.

The final goal of this research was to begin to identify which outcome expectancies differentiate youth who gamble with no associated difficulties from those who are developing or have gambling problems. Therefore, for youth participating in gambling activities, direct logistic regression analysis was performed using the scales of the GEQ to predict group membership: social gambler (DSM-IV-MR-J = 0–1) or problem gambler (at-risk gamblers and PPGs, DSM-IV-MR-J = 2–9). Direct logistic regression was undertaken to evaluate the contribution made by each predictor over and above that of the other predictors (Tabachnick & Fidell, 1996). Given that the criterion variable, group membership, is dichotomous and that the distributions of the independent variables (the five scales of the GEQ) are not likely to satisfy the assumptions of normality, logistic regression analysis is preferred to discriminant analysis (Tabachnick & Fidell, 1996). It should be noted that when used with dichotomous variables, like diagnostic categories, discriminant analysis tends to overestimate the magnitude of association (Davis & Offord, 1997) and may lead to the inclusion of too many predictor variables in the regression equation.

Prevalence findings

Of the total adolescent sample, 70.3% reported having gambled with money over the past 12 months. Of those participants who reported gambling, more males (82.4%) reported gambling than females (61.3%). Based upon gambling behaviour and the DSM-IV-MR-J criteria, overall, 5.0% of youth met the criteria for probable pathological gambling (scores of ≥ 4), 10.9% of the sample were considered at risk for pathological gambling (scores of 2–3), and 54.4% were considered to be social gamblers (scores of 0–1). More males gambled than females, and they also exhibited a higher prevalence of gambling-related problems: the rates for probable pathological gambling (9.3%) and at-risk gambling (18.3%) among males were greater than those for females (1.9% and 5.3%, respectively). Similarly, the rates of probable pathological gambling (6.5%) and at-risk gambling (11.5%) among older adolescents were higher than those for younger adolescents (2.8% and 9.7%, respectively). Gambling participation rates are reported in Table 1.

Table 1.
Gambling participation rates (past year) for the total sample

An independent samples t-test was performed to test for age differences across gender. Although the mean difference of .12 was statistically significant [t(953) = 3.82, p < .05], its clinical meaningfulness is questionable, as it is most likely attributable to the large sample size of the study.

Factorial ANOVA among DSM gambling groups, gender, and age groups

Significant main effects of gambling severity were found on all scales of the GEQ: enjoyment/arousal [F(3, 986) = 23.29, p < .01, partial η² = .066 ], self-enhancement [F(3, 986) = 5.70, p < .01, partial η² = .017], money [F(3, 986) = 18.34, p < .01, partial η² = .053], overinvolvement [F(3, 986) = 4.99, p < .01, partial η² = .015], and emotional impact [F(3, 986) = 26.21, p < .01, partial η² = .074].

On each of the three positive expectancy scales, PPGs and at-risk gamblers endorsed items on the enjoyment/arousal, self-enhancement, and money scales more highly than social gamblers and non-gamblers. Similarly, social gamblers endorsed the enjoyment/arousal and money scales more positively than non-gamblers. In terms of negative expectancies, non-gamblers endorsed the emotional impact scale more highly than social gamblers, at-risk gamblers, and PPGs; non-gamblers also endorsed the overinvolvement scale more highly than social gamblers. PPGs differed significantly from social gamblers and at-risk gamblers in their endorsement of the overinvolvement scale. Mean scores of the Dunnett's C Post Hoc results are summarized in Table 2.

Table 2.
DSM gambling group differences on the five scales of the GEQ

*p < .01
Range of scores: 1–7

A significant main effect of gender was found for enjoyment/arousal [F(1, 986) = 16.89, p < .01, partial η² = .017], money [F(1, 986) = 12.28, p < .01, partial η² = .012], and emotional impact [F(1, 986) = 16.74, p < .01, partial η² = .017]. Males were found to have endorsed the two positive expectancy scales, enjoyment/arousal and money, more positively than females. On the negative expectancy scale of emotional impact, however, females reported higher scores than males (see Table 3 for the means for both males and females on all scales).

Table 3.
Gender differences on the GEQ

*p < .01
Range of scores: 1–7

Developmentally, statistically significant differences were found among adolescents for enjoyment/arousal [F(1, 986) = 8.94, p < .01, partial η² = .009] and emotional impact [F(1, 986) = 12.58, p < .01, partial η² = .013]. Older adolescents endorsed the positive expectancy scale of enjoyment/arousal more highly than younger adolescents, who were more perceptive of the negative outcome of emotional impact (see Table 4 for age differences).

Table 4.
Developmental differences on the GEQ

* p < .01
Range of scores: 1–7

A significant interaction between gender and age was found on the enjoyment/arousal scale [F(1, 986) = 20.73, p < .01, partial η² = .021]. A significant difference was found between female adolescents aged 11–14 years and those aged 15–18 years. Older females (M = 4.61) endorsed items significantly more highly on the enjoyment/arousal scale than younger females (M = 3.82).

Logistic regression analyses

Direct logistic regression was used to identify which combination of scales of the GEQ best predicts category membership; social gambler or problem gambler. Separate direct logistic regression analyses were performed for males and females because of their distinct behavioural characteristics. For these analyses, the DSM criteria for social gamblers and problem gamblers (i.e., at-risk gamblers and PPGs) served as the criterion variable while four of the five GEQ scales and the age variable (two levels: 11–14, 15–18) were used as the predictor variables. In keeping with the previous univariate analyses, in which there were no significant differences found among social gamblers, at-risk gamblers, and PPGs mean scores on the emotional impact scale, the emotional impact variable was considered unrelated to the dependent variable of problem gambling group membership and was therefore not included in the logistic regression analyses discussed here. Age was included in the analysis because some developmental differences were observed in the univariate analyses. The age variable was entered into the analysis as its own block (block 1), while the remaining predictor variables were entered simultaneously into the logistic regression analysis as block 2. A less stringent criterion for significance was used, in the range of .05 to .10, as recommended by Hosmer and Lemeshow (1989).

For males, the results of the direct logistic regression indicated that the GEQ scales of enjoyment/arousal, self-enhancement, money, and overinvolvement all significantly contribute to the prediction model. The Hosmer–Lemeshow goodness-of-fit statistic indicated that the model fit was adequate (χ²(8, N = 354) = 9.12, p = .33). The contribution of each of the predictors is summarized in Table 5.

Table 5.
Direct logistic regression predicting gambling severity among male gamblers

Emotional impact was not included in the analysis.

In the prediction model, expectancies of enjoyment/arousal proved to be the strongest predictor: an increment of 1 on the enjoyment/arousal scale results in that individual being 1.6 times more likely to be a problem gambler. Similar increments on the money and self-enhancement scales are associated with males being 1.5 and 1.3 times (respectively) more likely than their peers to be problem gamblers. High scores on the negative expectancy scale of overinvolvement also served as a predictor of problem gambling, with males endorsing overinvolvement as a probable outcome being 1.3 times more likely to be problem gamblers. The resulting logistic regression equation classified 72% of cases correctly. It should be noted that this is a marginal increase in the overall classification rate (66%) had all of the gamblers been classified as social gamblers. Therefore, of greatest significance is the number of problem gamblers correctly classified; 39% of problem gamblers (n = 46) were predicted using these four scales (see Table 6).

Table 6.
Classification table for direct logistic regression model for male gamblers

Social gambler = social gambler on DSM-IV-MR-J (scores 0–1)
Problem gambler = at-risk gamblers + PPGs on DSM-IV-MR-J (scores ≥ 2)

The analysis was repeated for females, and the results of this direct logistic regression are presented in Table 7.

Table 7.
Direct logistic regression predicting gambling severity among female gamblers

Emotional impact was not included in this analysis.

For females, expectancies of enjoyment/arousal and money were the only significant predictors of gambling group membership within the model. An increment of 1 on both the enjoyment/arousal and money scales resulted in females being 1.4 times more likely to belong to the problem gambling group. The Hosmer–Lemeshow goodness-of-fit statistic was nonsignificant (χ²(8, N = 351) = 7.80, p = .45), suggesting adequate goodness-of-fit. Despite 88% of the cases being classified correctly, however, this logistic regression model resulted in all problem gamblers being inappropriately classified (see Table 8). Therefore, for females, the predictive value of outcome expectancies is very low.

Table 8.
Classification table for direct logistic regression for female gamblers