Introduction
Adolescent risk-taking behaviours are well-established causes of illness and injury (Turner et al.
2004). ‘Problem Behaviour Theory’ is foundational to modern adolescent risk research, and suggests that adolescents develop and exhibit risk-taking behaviours in related groups based on a variety of upstream determinants (called risk factors) (Jessor
1991,
2014). Such risk factors lead to the development of multiple risk behaviours that cluster together in predictable patterns within populations of young people (Pickett et al.
2002; De La Haye et al.
2014; Pfortner et al.
2015). Research and ongoing surveillance efforts in the field of adolescent risk-taking should, therefore, focus on these behaviours, both individually and in composite, to inform prevention efforts aimed specifically at adolescents. While other studies consider “risk behaviours” in the context of intent or motivation (Pérez Fuentes et al.
2016), this study uses the term to define the true action itself that poses potential harm to adolescents (e.g., Pickett et al.
2002; Riesch et al.
2013).
Over 2 decades ago, the US Centers of Disease Control and Prevention (CDC) created a risk behaviour framework that classified adolescent risk behaviours using evidence derived from the US Youth Risk Behaviour Survey (Kann et al.
2016). This long-standing framework categorizes risk behaviours across six domains that are associated with leading causes of morbidity and mortality among American youths: tobacco use, alcohol and illicit substance use, high-risk sexual behaviour, injury-prone and violent behaviour, unhealthy dietary patterns, and physical inactivity (Kann et al.
2016). These domains, informed by peer-reviewed literature (Brener et al.
2004), were established by consensus and are regularly updated to include emergent types of adolescent risk behaviours. Although this framework was initially developed for surveillance and policy development, it is probably the most frequently applied tool for population-based and clinical activities because of its broad consideration of varying behaviours associated with adolescent health.
There is a rich literature available that describes inter-relationships between adolescents’ risk behaviours and their potential effects on a variety of negative health outcomes (Yarber and Parrillo
1992; Lytle
2002; Schane et al.
2010; Spring et al.
2012; Thompson et al.
2014); however, such studies rarely consider the full complement of behaviours outlined within the CDC framework. This is particularly true in our own country of Canada. As a consequence, there may be an incomplete conceptual understanding of contemporary adolescent risk-taking behaviour and how such behaviours tend to develop and occur concurrently. For example, a large body of literature on adolescent risk-taking focuses on behaviours found in the stereotypical delinquent adolescent (i.e., the CDC domains of: alcohol and illicit substance use, tobacco use, and high-risk sexual behaviours) (Lindberg et al.
1995; Turner et al.
2004). Although these behaviours are suggestive of a high-risk lifestyle, they may also be related conceptually (and mathematically) to risk behaviours found in other domains within the CDC framework. In addition, there may be new risk behaviours within the CDC domains reflective of more contemporary patterns [e.g., e-cigarettes in tobacco use, and caffeinated energy drinks in alcohol and illicit substance use (Seifert et al.
2011; Goniewicz et al.
2016)], that are not included in these traditionally defined risk behaviour clusters. Behaviours from those and other domains in the CDC framework may be inter-related in different ways that reveal new patterns of risk behaviour. Finally, it is possible that adolescents in Canada engaged in risk-taking in ways that are unique from their American counterparts, and most of the existing empirical research in this field has been concentrated in the United States (Basen-Engquist et al.
1996; Riesch et al.
2013).
We had the opportunity, through analyses that involved both exploratory and confirmatory methods and a large population-based study of Canadians adolescents (Currie et al.
2012), to perform a contemporary analysis of adolescent risk-taking in Canada. Our objective was to explore inter-relationships between contemporary expressions of adolescent risk-taking, yet inspired by the long-standing CDC framework, and as a result, to create and validate new composite indicators of adolescent risk behaviours in a Canadian adolescent population. Our hope was that this would provide valuable new information in support of preventive initiatives in our country, and perhaps elsewhere.
Results
Sample population
Of the 30,096 responses available for study, 13,806 were in Grades 9–10 and 16,290 were in Grades 6–8. The proportion of students identified as being in the high-risk category for each of the risk behaviours included in the final model can be found in Table
2.
Table 2
Students in each risk level for all risk behaviours in the final exploratory model (Canada 2014)
Grades 9–10 students |
CDC domain 1: smoking cigarettes |
Lifetime smoking history | 10,682 | (80.7) | 1759 | (13.3) | 798 | (6.0) | 584 |
Use of alternative tobacco products | 9784 | (74.1) | 1896 | (14.4) | 1519 | (11.5) | 624 |
CDC domain 2: alcohol and illicit substance use |
Frequency of alcohol consumption | 5437 | (40.8) | 4192 | (31.5) | 3701 | (27.8) | 492 |
Number of drinks per typical event | 6022 | (45.8) | 2704 | (20.6) | 4409 | (33.6) | 688 |
Lifetime drunkenness history | 8676 | (65.3) | 3139 | (23.6) | 1465 | (11.0) | 543 |
Binge drinking | 8452 | (66.1) | 2968 | (23.2) | 1363 | (10.7) | 1040 |
Illicit drug use | 10,256 | (77.7) | 594 | (4.5) | 2351 | (17.8) | 622 |
Lifetime cannabis use | 10,082 | (76.5) | 793 | (6.0) | 2303 | (17.5) | 646 |
CDC domain 3: high-risk sexual behaviours |
Sex and contraceptive use | 7503 | (77.7) | 1828 | (18.9) | 329 | (3.4) | 4164 |
CDC domain 4: high-risk manifest behaviours |
Physical fighting | 10,199 | (75.7) | 1558 | (11.6) | 1715 | (12.7) | 352 |
Non-helmet use on a bicycle | 4961 | (39.4) | 3025 | (24.1) | 4593 | (36.5) | 1244 |
CDC domain 5: unhealthy dietary pattern |
Sweet consumption | 4601 | (34.5) | 4510 | (33.9) | 4211 | (31.6) | 502 |
Sugar-sweetened soda consumption | 7868 | (58.0) | 2943 | (21.7) | 2747 | (20.3) | 264 |
Chip consumption | 9508 | (70.3) | 2646 | (19.6) | 1374 | (10.2) | 295 |
Caffeinated energy drink consumption | 9921 | (72.5) | 2848 | (20.8) | 911 | (6.7) | 143 |
Low fruit consumption | 6234 | (45.5) | 5928 | (43.3) | 1542 | (11.3) | 119 |
Low vegetable consumption | 6042 | (44.4) | 5867 | (43.1) | 1695 | (12.5) | 219 |
Low orange vegetable consumption | 2096 | (15.4) | 6174 | (45.3) | 5351 | (39.3) | 203 |
CDC domain 6: physical inactivity |
Watching TV or videos | 3619 | (28.6) | 3885 | (30.7) | 5151 | (40.7) | 1168 |
Playing video games | 5282 | (41.8) | 2599 | (20.6) | 4750 | (37.6) | 1193 |
Using the computer | 3847 | (30.7) | 2603 | (20.8) | 6073 | (48.5) | 1301 |
Outdoor play | 3184 | (25.5) | 5086 | (40.8) | 4199 | (33.7) | 1355 |
Organized sport | 4127 | (31.3) | 5341 | (40.5) | 3730 | (28.3) | 626 |
Outdoor exercise | 5605 | (42.2) | 3747 | (28.2) | 3926 | (29.6) | 547 |
Grades 6–8 students |
CDC domain 1: smoking cigarettes |
Lifetime smoking history | 14,730 | (94.6) | 626 | (4.0) | 216 | (1.4) | 723 |
Use of alternative tobacco products | 14,342 | (92.5) | 745 | (4.8) | 417 | (2.7) | 790 |
CDC domain 2: alcohol and illicit substance use |
Frequency of alcohol consumption | 11,050 | (70.7) | 3278 | (21.0) | 1293 | (8.3) | 673 |
Lifetime drunkenness history | 14,290 | (91.9) | 1046 | (6.7) | 212 | (1.4) | 746 |
CDC domain 4: high-risk manifest behaviours |
Physical fighting | 10,717 | (68.2) | 2285 | (14.5) | 2715 | (17.3) | 578 |
Non-helmet use on a bicycle | 7576 | (48.6) | 4382 | (29.3) | 3320 | (22.2) | 1316 |
CDC domain 5: unhealthy dietary pattern |
Sweet consumption | 6339 | (40.9) | 4666 | (30.1) | 4489 | (29.0) | 801 |
Sugar-sweetened soda consumption | 10,550 | (66.4) | 3040 | (19.1) | 2297 | (11.5) | 408 |
Chip consumption | 11,569 | (72.9) | 2693 | (17.0) | 1612 | (10.2) | 420 |
Caffeinated energy drink consumption | 13,285 | (82.8) | 2153 | (13.4) | 601 | (3.7) | 255 |
Low fruit consumption | 8557 | (53.2) | 6189 | (38.5) | 1345 | (8.4) | 203 |
Low vegetable consumption | 7674 | (48.4) | 6277 | (39.6) | 1901 | (12.0) | 443 |
Low orange vegetable consumption | 2959 | (18.5) | 7131 | (44.7) | 5868 | (36.8) | 336 |
CDC domain 6: physical inactivity |
Watching TV or videos | 5449 | (36.4) | 4463 | (29.8) | 5043 | (33.7) | 1338 |
Playing video games | 6769 | (45.3) | 3473 | (23.3) | 4696 | (31.4) | 1357 |
Using the computer | 7161 | (48.2) | 3035 | (20.4) | 4668 | (31.4) | 1430 |
Outdoor play | 4572 | (30.8) | 6249 | (42.1) | 4008 | (27.0) | 1464 |
Organized sport | 5377 | (34.7) | 6702 | (43.2) | 3427 | (22.1) | 788 |
Outdoor exercise | 8020 | (50.3) | 4331 | (27.1) | 3607 | (22.6) | 337 |
Exploratory and confirmatory factor analysis
After consideration of the available 28 risk behaviours, a three-factor solution emerged from the exploratory factor analyses within both grade groups. The final model in both grade groups had an independent cluster solution. Eigenvalues for all common factors in the models were above the 90% confidence intervals from the parallel analyses, suggesting that variances explained by the factors were better than a chance finding. Final eigenvalues for the Grades 9–10 model were 5.08, 2.01, and 1.13 (N = 3594). For Grades 6–8 students, findings were consistent; the final eigenvalues were 3.06, 1.65, and 1.08 (N = 5586). Only students with responses to all risk behaviours were included in the factor analyses. Given the reduced sample sizes used for the final exploratory factor analyses due to missing responses, sensitivity analyses using full information maximum-likelihood imputation were performed, and no significant changes to factor structure, eigenvalues, or factor loadings were identified.
Based on similar results from the exploratory analyses for both grade groups, the three risk behaviour categories were labeled together based on a general conceptual understanding of the behaviours that emerged from their respective exploratory factor analyses (Tables
3,
4). The first common factor showed behaviours associated with substance use and externalizing risk-taking such as fighting, non-helmet use while riding on a bicycle, and risky sexual behaviour. We called this category ‘
Overt Risk Taking’. The second factor identified behaviours associated with low consumption of nutritious food (such as fruits and vegetables) and low participation in different forms of moderate-to-vigorous physical activity (such as organized sports and free play). Because of the omission of behaviours associated with healthy, balanced lifestyles, we called this category ‘
Aversion to a Healthy Lifestyle’. The third factor grouped sedentary screen time activities together with unhealthy snacking behaviours (i.e., potato chip and soda consumption)—we called this category the ‘
Screen Time Syndrome’. Cronbach’s alpha values calculated for each of the risk behaviour categories in both grade groups were all above 0.60 suggesting acceptable levels of internal consistency.
Table 3
Exploratory factor analysis on risk behaviours considering all domains in the US Centers for Disease Control and Prevention risk framework in Grades 9–10 students (N = 3594) (Canada 2014)
CDC domain 1: smoking cigarettes |
Lifetime smoking history | 0.65 | | | 0.062 |
Use of alternative tobacco products | 0.71 | | | 0.048 |
CDC domain 2: alcohol and illicit substance use |
Frequency of alcohol consumption | 0.76 | | | 0.080 |
Number of drinks per typical event | 0.80 | | | 0.096 |
Lifetime drunkenness history | 0.85 | | | 0.065 |
Binge drinking | 0.81 | | | 0.061 |
Illicit drug use | 0.33 | | | 0.084 |
Lifetime cannabis use | 0.73 | | | 0.050 |
CDC domain 3: high-risk sexual behaviours |
Sex and contraceptive use | 0.46 | | | 0.045 |
CDC domain 4: high-risk manifest behaviours |
Non-helmet use on a bicycle | 0.32 | | | 0.077 |
CDC domain 5: unhealthy dietary pattern |
Caffeinated energy drink consumption | 0.37 | | | 0.022 |
CDC domain 5: unhealthy dietary pattern |
Fruit consumption | | 0.63 | | 0.040 |
Vegetable consumption | | 0.56 | | 0.039 |
Orange vegetable consumption | | 0.56 | | 0.020 |
CDC domain 6: physical inactivity |
Duration of outdoor play | | 0.39 | | 0.041 |
Participation in organized sports | | 0.48 | | 0.024 |
Frequency of exercise outside school hours | | 0.47 | | 0.026 |
CDC domain 5: unhealthy dietary pattern |
Sweet consumption | | | 0.39 | 0.011 |
Sugar-sweetened soda consumption | | | 0.59 | 0.036 |
Chip consumption | | | 0.51 | 0.015 |
CDC domain 6: physical inactivity |
Watching TV or videos | | | 0.38 | 0.014 |
Playing video games | | | 0.44 | 0.014 |
Using a computer | | | 0.31 | 0.018 |
Final eigenvalues | 5.08 | 2.01 | 1.13 | |
Cronbach’s alpha (standardized) | 0.87 | 0.67 | 0.61 | |
McDonald’s omega | 0.88 | 0.54 | 0.57 | |
Confirmatory factor analysisa: RMSEA (90% CI) | 0.088 (0.086, 0.089) |
Confirmatory factor analysisa: SRMR | 0.071 | |
Confirmatory factor analysisa: AGFI | 0.807 | |
Table 4
Exploratory factor analysis on risk behaviours considering all domains in the US Centers for Disease Control and Prevention risk framework in Grades 6–8 students (N = 5586) (Canada 2014)
CDC domain 1: smoking cigarettes |
Lifetime smoking history | 0.73 | | | 0.047 |
Use of alternative tobacco products | 0.76 | | | 0.067 |
CDC domain 2: alcohol and illicit substance use |
Frequency of alcohol consumption | 0.57 | | | 0.074 |
Lifetime drunkenness history | 0.65 | | | 0.050 |
CDC domain 4: high-risk manifest behaviours |
Non-helmet use on a bicycle | 0.35 | | | 0.149 |
Physical fighting | 0.30 | | | 0.029 |
CDC domain 5: unhealthy dietary pattern |
Caffeinated energy drink consumption | 0.42 | | | 0.046 |
CDC domain 5: unhealthy dietary pattern |
Fruit consumption | | 0.62 | | 0.045 |
Vegetable consumption | | 0.61 | | 0.053 |
Orange vegetable consumption | | 0.51 | | 0.028 |
CDC domain 6: physical inactivity |
Duration of outdoor play | | 0.36 | | 0.052 |
Participation in organized sports | | 0.38 | | 0.034 |
Frequency of exercise outside school hours | | 0.48 | | 0.029 |
CDC domain 5: unhealthy dietary pattern |
Sweet consumption | | | 0.37 | 0.023 |
Sugar-sweetened soda consumption | | | 0.50 | 0.066 |
Chip consumption | | | 0.43 | 0.034 |
CDC domain 6: physical inactivity |
Watching TV or videos | | | 0.50 | 0.029 |
Playing video games | | | 0.55 | 0.029 |
Using a computer | | | 0.50 | 0.043 |
Final eigenvalues | 3.06 | 1.65 | 1.08 | |
Cronbach’s alpha (standardized) | 0.75 | 0.65 | 0.66 | |
McDonald’s omega | 0.68 | 0.63 | 0.63 | |
Confirmatory factor analysisa: RMSEA (90% CI) | 0.074 (0.072, 0.076) | |
Confirmatory factor analysisa: SRMR | 0.067 | |
Confirmatory factor analysisa: AGFI | 0.885 | |
In both grade groups, modest correlations were observed between common factors.
Overt Risk Taking was marginally correlated to both
Aversion to a Healthy Lifestyle (Grades 9–10:
r = 0.06,
r
corrected = 0.08; Grades 6–8:
r = 0.21,
r
corrected = 0.30) and the
Screen Time Syndrome (Grades 9–10:
r = 0.14,
r
corrected = 0.19; Grades 6–8:
r = 0.21,
r
corrected = 0.32). There were positive correlations of moderate strength between the
Screen Time Syndrome and
Aversion to a Healthy Lifestyle in both grade groups (Grades 9–10:
r = 0.30,
r
corrected = 0.47; Grades 6–8:
r = 0.31,
r
corrected = 0.47). However, overall these correlations suggest that these three factors are distinct from one another. Intraclass correlations (ICC) calculated for each risk behaviour were variable and suggested low-to-moderate variance attributable to the school level (Grades 9–10: ICC range = 0.011–0.096, Table
3; Grades 6–8: ICC range = 0.023–0.149, Table
4).
Confirmatory factor analyses suggested that the Grades 9–10 (SRMR = 0.071, RMSEA = 0.088, AGFI = 0.807,
N = 3693;
χ
2 = 6660,
df = 227,
p < 0.001) and the Grades 6–8 final models (SRMR = 0.067, RMSEA = 0.074, AGFI = 0.885,
N = 5984;
χ
2 = 4995,
df = 149,
p < 0.001) had modest fits (Hooper et al.
2008).
Confirmatory factor analyses were also performed to investigate a two-factor model that combined items from the Aversion to a Healthy Lifestyle and Screen Time Syndrome categories. This two-factor model performed poorer across all of the fit indices (Grades 9–10: SRMR = 0.079, RMSEA = 0.094, AGFI = 0.782; Grades 6–8: SRMR = 0.083, RMSEA = 0.094, AGFI = 0.812) when compared to the three-factor model.
Discussion
This study provides a contemporary and comprehensive examination of risk-taking behaviours among Canadian adolescents—a population that is vital to examine in the field of multiple risk behaviour as it is a critical period of the life course. The objective of this study was to evaluate relationships between risk behaviours, both novel and contemporary, derived from a list inspired by the diverse domains of the CDC framework. We conducted this analysis within a Canadian context. Based on that objective, we found that three composite indicators of clustered risk behaviour emerged from our analysis. Interestingly, these new latent constructs encompassed behaviours that crossed all six domains of risk described within the CDC framework. Our findings were fairly consistent across two broad developmental periods (Grades 6–8 and 9–10), although the items available to measure adolescent risk-taking were limited in the youngest age group.
The CDC’s Youth Risk Behaviour Survey of adolescent risk behaviour evaluates and monitors the domains of behaviour most closely associated with known leading causes of morbidity and mortality (Kann et al.
2016). Continuous revision based on public health data and expert opinion means that it consistently captures a comprehensive list of new and long-standing risk behaviours that impact youth, which makes it a frequently used tool for prevention and harm-reduction programs. However, many current public health programs remain outdated in their use of the CDC framework by targeting individual domains of behaviour (i.e., alcohol and illicit substance use), or even individual risk behaviours within a domain (i.e., alcohol consumption) for public health interventions and ignore the well-established concept of their inter-related and clustered natures. By incorporating lessons of Problem Behaviour Theory (Jessor
1991,
2014), our study uses the comprehensive CDC risk domains to identify three clusters of risk behaviour that may provide more focused public health interventions targeting contemporary populations of young Canadians.
Our approach to conceptualizing risk behaviour recognizes the complex relationships that exist amongst them, as well as their possible joint effects on disease etiology. The items indicated by the CDC domains are intimately related in interpretable and potentially unexpected ways, consistent across grade groups. As a result, these behaviours should be observed and measured collectively under each of the three categories to be properly understood and managed.
The three risk categories that were identified by factor analysis incorporated items that are well recognized within the adolescent health research literature. The
Overt Risk Taking category largely encompasses behaviours found in the traditional adolescent risk studies (Maggs et al.
1997; de Looze et al.
2012). However, we believe that the benefits of our analysis lie in the more contemporary expressions of risk-taking that were incorporated. For example, the inclusion of
caffeinated energy drinks and
alternative tobacco products highlights emergent areas of related risk that reveal either the true breadth of this category, or behaviours associated with more moderate risk tolerance that indicate early development of risk-taking in this domain. In contrast, the traditional public health programs targeting individual behaviours, such as cigarette use, may forego the opportunity to educate and prevent other behaviours within the same category that an adolescent is likely already participating in (such as consumption of caffeinated energy drinks, alternative tobacco products, cannabis use, and riding a bicycle without a helmet). The
Screen Time Syndrome and
Aversion to a Healthy Lifestyle categories indicated close relationships between diet, physical activity, and sedentary behaviour that have been recognized in the past research (Leech et al.
2014). Nevertheless, our empirical distinction between these categories also supports studies that show that a lack of moderate-to-vigorous physical activity and excessive screen time represent separate and distinct behaviours among adolescent populations (Pearson et al.
2014; Brindova et al.
2015). Our study’s risk categories represent measures of three separate types of adolescent risk behaviour that were found to be robust through confirmatory analysis in a separate subset of our study population.
Although psychometric research on the relationships amongst risk behaviours has been conducted in the past, none have used an established framework to ensure that they have captured a group of behaviours that are associated with the current leading causes of illness and injury among youths. Our three categories incorporate behaviours that span each of the six of the CDC risk behaviour domains, and the resultant composite scores are, perhaps, more consistent with the way that adolescents behave socially compared to other studies of multiple risk behaviours. These three risk categories have implications for the development and targeting of public health interventions that improve upon individual risk behaviour approaches, and can broaden clustered risk behaviour approaches with a narrow scope (i.e., Sloboda et al.
2009).
Of importance, our analysis extends the existing research (e.g., Pickett et al.
2002; De La Haye et al.
2014) by being amongst the first studies to factor analytically derive composite indicators of risk-taking based on an established framework. This study identified three ways that adolescents engage in risk-taking behaviours—each presumably having their own upstream determinants and downstream health consequences. Further research is now needed to confirm these as stable and consistent composite indicators of risk-taking in other study populations and contexts, and to evaluate the health outcomes and risk factors associated with each distinct category. Future intervention programs could then target the risk factors of each category to address their associated negative health outcomes (Jackson et al.
2012; Hale et al.
2014).
Admittedly, research such as this is often limited by its reliance on self-reported data. The HBSC attempts to minimize this limitation through the emphasis of confidentiality of responses (Currie et al.
2012). Nevertheless, students may not have answered truthfully to all the questions due to social desirability biases. Similarly, risk behaviour and sensitive questions have higher rates of non-response, most notably those surrounding sexual behaviour. Several of the items in the HBSC ask about days of lifetime exposure to specific risk behaviours and may misclassify newly emergent high-frequency engagement as moderate engagement. Finally, the CDC risk framework that inspired our work may not be completely applicable to the Canadian HBSC study population, based on cultural and age differences. Finally, we reported the intraclass correlation for each risk behaviour included in our composite measures and note that some behaviours showed moderate clustering effects at the school level. Although the majority of the risk behaviours in our study had negligible clustering effects at the school level, this analysis did not account for such clustering and may have overestimated variance at the student-level. Based on our reported measures, future analyses may choose to account for school-level clustering effects.
Conclusion
This study used a large sample of Canadian adolescents to evaluate relationships amongst adolescent risk behaviours. This psychometric research was inspired by the six-domain framework outlined by the CDC (Kann et al.
2016). Our empirical analysis, which included both exploratory and confirmatory factor analytic techniques, found that adolescent risk behaviours cluster in predictable patterns crossing the different CDC risk domains. Three categories of risk behaviours emerged based on the six-domain framework: (1)
Overt Risk Taking, (2)
Aversion to a Healthy Lifestyle, and (3)
Screen Time Syndrome. These categories build on the existing studies of multiple risk behaviour, and inform research and intervention efforts aimed at preventing adolescent illness and injury. Future research could use this new framework of adolescent risk behaviour to study their upstream determinants, as well as their joint causes of negative health outcomes.
Compliance with ethical standards