Are experimental smokers different from their never-smoking classmates? A multilevel analysis of Canadian youth in grades 9 to 12 - CDIC: Vol 34, No 2-3, July 2014

Volume 34 · Number 2-3 · July 2014

Are experimental smokers different from their never-smoking classmates? A multilevel analysis of Canadian youth in grades 9 to 12

S. C. Kaai, PhD (1); S. R. Manske, EdD (1, 2); S. T. Leatherdale, PhD (1); K. S. Brown, PhD (1, 2, 3); D. Murnaghan, PhD (4)

https://doi.org/10.24095/hpcdp.34.2/3.07

This article has been peer reviewed.

Author references:

1. School of Public Health and Health Systems, University of Waterloo, Waterloo, Ontario, Canada
2. Propel Centre for Population Health Impact, University of Waterloo, Waterloo, Ontario, Canada
3. Department of Statistics and Actuarial Science, University of Waterloo, Waterloo, Ontario, Canada
4. School of Nursing, University of Prince Edward Island, Charlottetown, Prince Edward Island, Canada

Correspondence: Susan C. Kaai, School of Public Health and Health Systems, University of Waterloo, 200 University Avenue West, Waterloo, ON N2L 3G1; Tel.: 519-888-4567 ext. 31748; Fax: 519-746-8631; Email: skaai@uwaterloo.ca

Abstract

Introduction: Understanding the characteristics of experimental smoking among youth is critical for designing prevention programs. This study examined which student- and school-level factors differentiated experimental smokers from never smokers in a nationally representative sample of Canadian students in grades 9 to 12.

Methods: School-level data from the 2006 Canadian Census and one built environment characteristic (tobacco retailer density) were linked with data from secondary school students from the 2008–2009 Canadian Youth Smoking Survey and examined using multilevel logistic regression analyses.

Results: Experimental smoking rates varied across schools (p < .001). The location (adjusted odds ratio [AOR] = 0.66, 95% CI: 0.49–0.89) of the school (urban vs. rural) was associated with the odds of a student being an experimental smoker versus a never smoker when adjusting for student characteristics. Students were more likely to be experimental smokers if they were in a lower grade, reported low school connectedness, used alcohol or marijuana, believed that smoking can help people relax, received pocket money each week and had a family member or close friend who smoked cigarettes.

Conclusion: School-based tobacco prevention programs need to be grade-sensitive and comprehensive in scope; include strategies that can increase students' attachment to their school; and address multi-substance use, tobacco-related beliefs and the use of pocket money. These programs should also reach out to students who have smoking friends and family members. Schools located in rural settings may require additional resources.

Keywords: tobacco smoking, youth, prevention, multilevel analysis, Canada

Introduction

Despite the proven harmful outcomes of smoking, youth smoking rates remain high in North America.^{Footnote 1} More than 16% and 20% of all annual deaths in Canada and the United States, respectively, result from tobacco-related diseases.^{Footnote 2,Footnote 5,Footnote 6} Lifetime smoking often commences as naive experimentation during adolescence and develops into a habit that is difficult to break.^{Footnote 7} Most adult smokers initiated smoking during their teenage years.^{Footnote 3} Evidence also suggests that adolescent smoking behaviour consists of distinct smoking trajectories or stages: susceptible never smokers, experimenters and established (or current) smokers.^{Footnote 8, Footnote 9, Footnote 10} Differentiating between these smoking stages is important to public health practitioners and educators who need to design prevention and intervention programs to match the risk and protective factors in these different stages.

A majority of studies examine established smoking stages.^{Footnote 10, Footnote 11, Footnote 12, Footnote 13, Footnote 14, Footnote 15, Footnote 16, Footnote 17} Considering that approximately three-quarters of students will experiment with smoking at least once before completing high school^{Footnote 18, Footnote 19} and that about one-third will become established smokers,^{Footnote 20} understanding the factors that differentiate experimental smokers from never smokers is critical to informing the development of the tobacco control programs designed to discourage students from experimenting with cigarettes.

Many researchers have used the Theory of Triadic Influence (TTI)^{Footnote 21} to understand the complex factors associated with adolescent smoking behaviour. TTI postulates that youth smoking behaviour is influenced by a combination of and interaction between intrapersonal, social context and broader societal influences. Intrapersonal risk factors associated with experimental smoking include age,^{Footnote 22} male sex,^{Footnote 23} use of alcohol or illicit drugs,^{Footnote 22} access to pocket money,^{Footnote 24} low school connectedness,^{Footnote 25} positive attitudes towards smoking^{Footnote 26} and perceiving clear school rules about smoking.^{Footnote 27} Existing social context influences include smoking family members^{Footnote 22, Footnote 23} or friends.^{Footnote 22, Footnote 27} The broader societal (or school-level) factors associated with experimental smoking include attending a school with a relatively high smoking rate in senior grades,^{Footnote 27} high density of tobacco retailers around the school^{Footnote 28} and living in a home that does not have a total ban on smoking.^{Footnote 29} Chan and Leatherdale^{Footnote 30} explored the relationship between tobacco retailers and smoking susceptibility, occasional smoking and established smoking. They reported that the number of tobacco retailers located around a school was associated only with smoking susceptibility.^{Footnote 30} Other societal factors associated with established (not experimental) smoking include school location^{Footnote 31, Footnote 32, Footnote 33} and neighbourhood socioeconomic status (SES).^{Footnote 31, Footnote 32, Footnote 34, Footnote 35}

Nevertheless, there is a dearth of literature on the influence of school location (urban vs. rural), tobacco retailer density and the SES of the community around a school on students' experimental smoking when adjusting for other student-level factors. Because these school-level factors have previously been found to be associated with established smoking,^{Footnote 31, Footnote 32, Footnote 33, Footnote 34, Footnote 35, Footnote 36,Footnote 37,Footnote 38} we were interested in finding out whether these factors were also associated with experimental smoking among adolescents. As such, the purpose of this study was to examine which school neighbourhood and student-level characteristics differentiate experimental smokers from never smokers. Understanding these factors will provide new insight for public health practitioners and educators who develop smoking prevention strategies that effectively target youth in different stages of smoking. The authors have also written on the factors associated with current (or established) smoking.^{Footnote 39}

Methods

Design

The 2008–2009 Canadian Youth Smoking Survey (2008 YSS) is a nationally representative cross-sectional, school-based survey that is used to measure the determinants of youth smoking behaviour. It is a valid and reliable machine-readable, pencil and paper study.^{Footnote 40} (See Elton-Marshall et al.^{Footnote 41} and www.yss.uwaterloo.ca for detailed information on the survey development, design, survey weights and data collection protocol.) In brief, the target population consisted of all young Canadian residents in the appropriate grades attending public or private schools in all 10 provinces in Canada. The sample design was based on a stratified multistage design. The survey took about 20 to 30 minutes, and to ensure confidentiality, students placed completed questionnaires in an envelope that was sealed and placed in a larger classroom envelope. The University of Waterloo Office of Research and Ethics approved the survey methods.

Participants

The sample for this study was from the secondary school portion of 2008 YSS. This portion was administered to all sampled grade 9 to 12 students (n = 29 296) attending 133 schools from all 10 Canadian provinces. The student response rate was 73.2%.^{Footnote 41} Our study used only the subset of 18 072 students who were experimental or never smokers.

Data sources and measures

Outcome variables

Based on other research,^{Footnote 13,Footnote 28,Footnote 42,Footnote 43} we defined ''experimental smokers'' as those who had smoked in the last 30 days before the survey but had not smoked 100 cigarettes in their lifetime. This group was compared with ''never smokers,'' defined as those who reported never having smoked even a puff of a cigarette.^{Footnote 40}

Student (intrapersonal and social context) and school-level (broader societal) correlates

Selection of all variables was guided by TTI^{Footnote 21} and our literature review. We coded the intrapersonal factors (sex, grade, alcohol or marijuana use, pocket money, school connectedness, knowledge and attitude towards tobacco, and perception of school smoking rules) and social context measures (parents', siblings' and friends' smoking status) as listed in Table 1. Two school-level neighbourhood characteristics from the 2006 Canadian Census (i.e., location [urban vs. rural] and median household income, which is a proxy measure for school neighbourhood SES) were linked with the 2008 YSS dataset, as has been done by other researchers.^{Footnote 44,Footnote 45} Both school location and median household income data were derived from school postal codes using the Postal Code Conversion File that links between the postal code and Statistics Canada's standard 2006 Census geographical areas^{Footnote 46} (see Table 1). The 2008/09 Enhanced Points of Interest (EPOI) data file from Desktop Mapping Technologies Inc.^{Footnote 47} provided numerical data on the number of tobacco retailers located within a 1-kilometre radius of each school (Table 1).

Statistical data analyses

We used multilevel logistic regression to analyze the two-level nested data because it accounts for the clustering (interdependence) of students within schools by allowing the model intercept to vary across schools.^{Footnote 48} This produces accurate standard errors and reduces the likelihood of type 1 error.^{Footnote 49} Like other researchers,^{Footnote 27} we used a four-step modelling procedure. Model 1 is a null model computed to assess whether there was significant within-cluster interdependence to warrant the use of a multilevel approach. The main purpose for Model 2 was to determine the school-level variables that would directly affect the likelihood of a student being an experimental smoker rather than a never smoker. Model 3 used a random coefficient regression model to assess the strength of the direct effects of both the school- and student-level correlates.

Model 4 was developed to assess the contextual interactions between the school-level and student-level predictor variables. The SAS PROC GLIMMIX^{Footnote 50} procedure provided the initial estimates that were used in the PROC NLMIXED analysis for each model. Predictor variables that were not significant at p < .05 were removed until the final model only contained predictor variables that were significant at that p value. The intraclass correlation (ICC) measures the proportion of the total variance that occurs between schools. The σ²_m denotes the school-level variance, whereas the logistic distribution for the individual residual implies a variance of π²/3 = 3.29. This formula considers that the observed binary response actually represents a threshold continuous variable where 0 is observed below the threshold and 1 above.^{Footnote 48}

All analyses used SAS version 9.2 (SAS Institute Inc., Cary, NC, US).^{Footnote 50}

Results

Student- and school-level characteristics

Of the sample of grade 9 to 12 students, 16 044 (54.8%) were classified as never smokers and 2028 (6.9%) were classified as experimental smokers. The remainder were not included in our study. Boys made up 51% of the sample. The prevalence of experimental smoking did not differ by sex (x2 = 0.02; p = 0.89; dƒ = 1). With that exception, all other student characteristics tested were significant (p < .001).

The proportion of experimental smokers increased from grade 9 to 12; as the number of friends who smoke increased from 1 to 5; and as the amount of weekly pocket money increased (Table 2). The percentage of experimental smokers who used marijuana (36.8%) or alcohol (14.2%) was strikingly higher than the percentage of experimental smokers who did not use marijuana (2.4%) or alcohol (1%).

Of the total sample of 133 secondary schools, 69 were located in urban areas. The average experimental smoking rate among students in grades 9 to 12 in the 133 secondary schools was 6.2% (range, 0%–17.4%), and this was lower in urban schools (5.7%) than in rural schools (6.6%). The percentage of experimental smokers (11.1%; 1325/11 977) in urban schools did not significantly differ from that in rural schools (11.5%; 703/6095). The mean number of tobacco retailers within a 1-kilometre radius of the schools was 5.8 (standard deviation [SD] 10; range, 0–49 km). The mean household income within the postal code around each school was $56 424 (SD $14 574; range, $30 784–$97 706).

Multilevel analysis of experimental smoking

Table 3 shows results of the multilevel logistic regression analyses. The results from the null model (Model 1) showed a significant between-school random variation (Estimate [Standard Error (SE)] = 0.23 [0.05]; p < .001) in the likelihood of experimental smoking among grade 9 to 12 students. The estimates suggest that the school a student attends accounts for 6.5% of the variability in their likelihood of being an experimental smoker versus a never smoker. Model 2 results show that only school location was important, as students in urban schools were less likely to be experimental smokers than never smokers (adjusted odds ratio [AOR] = 0.74, 95% CI: 0.60–0.91) compared to students in rural schools. This neighbourhood characteristic explained 11.9% of the between-school variability in the likelihood of a student being an experimental smoker. The number of tobacco retailers within a 1-kilometre radius around a school was not associated (AOR = 0.99, 95% CI: 0.97–1.01) with experimental smoking. Additionally, the median household income that was used as a proxy measure for school neighbourhood SES was not associated (AOR = 0.93, 95% CI: 0.86–1.01) with the likelihood of a student being an experimental smoker versus a never smoker.

Model 3 identified the school-level characteristics that were significantly associated with the odds of a student being an experimental smoker when adjusting for student-level characteristics. When we first examined each of the three school-level variables separately (adjusting for the student-level variables), the location (AOR = 0.62, 95% CI: 0.46–0.82; urban vs. rural; data not shown) and the neighbourhood SES (AOR = 0.88, 95% CI: 0.79–0.98; data not shown) where schools were located were significantly associated with the odds of a student being an experimental smoker. However, when we put all the school-level (location, SES and number of tobacco retailers) and student-level variables in one final model, only school location (urban vs. rural) remained significant (AOR = 0.66, 95% CI: 0.49–0.89; see Table 3). None of the contextual interactions in Model 4 (results not shown) were associated with the outcome variable.

In summary, the final model suggests that there were no sex differences (AOR = 1.00, 95% CI: 0.86–1.16) in the likelihood of a student being an experimental smoker versus a never smoker. The odds of a student being an experimental smoker decreased when they attended an urban school (AOR = 0.66, 95% CI: 0.49–0.89) compared with a rural school. In terms of student-level findings, students who were in grade 10 and 11 were less likely to be experimental smokers than never smokers compared with those who were in grade 9 (grade 10 vs. grade 9: AOR = 0.75, 95% CI: 0.61–0.93; Grade 11 vs. grade 9: AOR = 0.71, 95% CI: 0.57–0.89). On the other hand, the odds of a student being an experimental smoker versus a never smoker increased with the amount of weekly pocket money they had to spend ($1–$20 vs. no pocket money: AOR = 1.59, 95% CI: 1.20–2.11; > $100 vs. no pocket money: AOR = 2.23, 95% CI: 1.66–2.99); a student's belief that smoking can help people relax (AOR = 3.37, 95% CI: 2.85–3.97); a student's perception that there are clear school rules on smoking (AOR = 1.56, 95% CI: 1.27–1.92); low school connectedness (AOR = 0.87, 95% CI: 0.83–0.92); and alcohol use (AOR = 3.51, 95% CI: 2.41–5.12). Marijuana use appeared to be very important as the odds of a student being an experimental smoker was more than 15 times higher (AOR = 15.4, 95% CI: 12.96–18.26) if the student reported using marijuana.

In terms of social context correlates, a student who reported that at least one parent (AOR = 1.29, 95% CI: 1.11–1.50) or sibling (AOR = 1.45, 95% CI: 1.22–1.73) smoked cigarettes was at an increased risk of being an experimental smoker versus a never smoker. In addition, the odds of a student being an experimental smoker increased significantly as the number of closest friends who smoke cigarettes increased, that is, AOR ranged from 3.69 (95% CI: 2.96–4.59) for one close friend versus no friends who smoke) to AOR of 10.52 (95% CI: 7.10–15.60) for four close friends versus no friends who smoke. Between schools variation was not accounted for by these student-level factors. School-to-school variation remained significant even after adjusting for student-level factors.

Discussion

Since some youth experience nicotine dependence within as little as a day after first inhaling cigarette smoke,^{Footnote 22} dissuading them from experimenting with cigarettes is an important way of preventing smoking. Our study identified four notable findings valuable to future tobacco control prevention programming. First, the likelihood of a student being an experimental smoker varied significantly across schools, a finding consistent with other research on youth smoking behaviour.^{Footnote 27,Footnote 51} This suggests that the characteristics of a student's school are associated with the likelihood that they will be an experimental smoker above and beyond the individual student's characteristics. Although school accounted for a modest 6.5% of the variability, when distributed across the broader secondary school population in Canada, it represents a notable amount of variation that cannot be ignored.

Second, our results supported TTI and also expanded on existing literatureFootnote 31, Footnote 32, Footnote 33, Footnote 34, Footnote 35 by showing that variables related to school location (i.e. rural vs. urban setting and the school neighbourhood SES [when analyzed alone]) were associated with experimental smoking after controlling for student-level characteristics. However, stronger and more in-depth studies would be necessary to help public health practitioners identify the specific characteristics in rural schools or schools located in low SES neighbourhoods that predispose students to experimental smoking.Footnote 31, Footnote 32, Footnote 33, Footnote 34, Footnote 35 Moreover, school location (urban vs. rural), neighbourhood SES and tobacco retailer density only explained part of the between-school variability; more surveillance activities are required to evaluate other types of school-level data such as linkages with the community and media and the role of school-based tobacco control programs and policies.Footnote 42, Footnote 52, Footnote 53

In contrast to other researchers' findings on experimental smoking,^{Footnote 28} we found that the number of tobacco retailers located around secondary schools was not associated with the outcome variable. This suggests that the number of tobacco retailers around a school is more important for those students who are susceptible to smoking^{Footnote 30} or established smokers^{Footnote 36, Footnote 37, Footnote 38, Footnote 39} than for students who are still experimenting with cigarettes. TTI variables drawn from the individual level may offer an explanation. Previous studies found that regular smokers were more likely to use retail sources, that is, buy cigarettes from tobacco retailers while experimenters used social sources such as ''borrowing'' cigarettes from friends or family, which made the location of retailers less important in our study.^{Footnote 54}

Third, the intrapersonal findings (i.e. grade, attitudes, pocket money, perception of anti-smoking rules, alcohol and marijuana use, school connectedness) from our study were consistent with existing literature.Footnote 21,Footnote 26 For example, students who reported pro-smoking attitudes, such as believing that smoking can help people relax, were more likely to be experimental smokers. This is not surprising; the TTI posits that adolescents' perceptions and beliefs represent the most proximal level of influence because they reflect the adolescents' ability to resist pressures to initiate and progress into advanced smoking behaviour.Footnote 21 The amount of pocket money students had available was also associated with experimental smoking, a finding consistent with that of Mohan et al.Footnote 24 Parents and guardians who give their adolescent children pocket money need to understand how this money is spent.

Our finding about students' perceptions of anti-smoking school rules is consistent with that of other researchers who indicated that tobacco control school policies or rules are not effective on their own but that suitable enforcement is necessary.Footnote 14,Footnote 16 Our study did not assess enforcement; however, plausible explanations include that existing smokers tend to notice anti-smoking policies relevant to them or that schools develop and implement policies in response to higher rates of tobacco use.Footnote 55 Perhaps the experimental smokers in our study reflect individual differences in oppositional defiant tendencies or sensation-seeking behaviour (we did not measure these characteristics), as reported in other studies.Footnote 56,Footnote 57

Our finding that alcohol use predicted experimental smoking is consistent with that of other studies.^{Footnote 22} Most striking was the finding that if a student reported using marijuana the odds of them being an experimental smoker (vs. a never smoker) were more than 15 times higher (AOR 15.4, 95% CI 12.96–18.26) than for those students who did not report usage. Although it is not possible to determine whether marijuana use precedes tobacco use or vice versa using our cross-sectional data, this finding highlights adolescent multi-substance or multi-risk behaviour and the importance of schools prioritizing the prevention of substance use (whether tobacco, marijuana, alcohol or combinations of substances) by optimizing limited resources through the use of multi-pronged strategies that target multiple substance use.^{Footnote 58,Footnote 59,Footnote 60,Footnote 61,Footnote 62,Footnote 63} This approach improves students' educational outcomes and also encourages healthy social behaviours that help students resist substance abuse and feel more connected to their school.^{Footnote 63,Footnote 64,Footnote 65} Consistent with other research,^{Footnote 25,Footnote 66} our results show that students who feel more connected to their school are less likely to initiate risky behaviour such as tobacco use. It is also consistent with current efforts in Canada (e.g. Pan-Canadian Joint Consortium for School Health^{Footnote 64} and New Brunswick Wellness Strategy^{Footnote 65}) to address ''upstream'' issues in school settings to create healthy environments and provide skills to enable youth to resist any form of substance use.

Fourth, the results about friends and family who smoke are consistent with existing evidence^{Footnote 22,Footnote 23,Footnote 27} and support the TTI,^{Footnote 21} which posits that this group forms the immediate social environment that contributes to the social pressure (e.g. by reinforcing the behaviour through offering cigarettes or modelling smoking) on adolescents to experiment with tobacco. The implication for this finding is that school-based tobacco control programs should equip students with the necessary information and skills to deal with any form of pressure that may predispose them to experimental smoking.^{Footnote 8}

That students in grades 10 and 11 were less likely to be experimental smokers (vs. never smokers) than those who in grade 9 was consistent with results of studies that examined established smoking.^{Footnote 20} Students in higher grades may have moved on from experimental smoking to regular or established smoking. This suggests that school-based prevention strategies should be implemented early, in elementary school, and sustained into high school and postsecondary years (subject to availability of resources). Unlike one Chinese study,^{Footnote 23} we did not find sex to be associated with the outcome variable. However, although boys did not differ from girls in our analyses, their decisions to experiment with smoking may have different influences,^{Footnote 67} and to the extent that this is true, school-based interventions may still have to consider sex.

Strengths and limitations

The strengths of this study include our use of nationally representative data of Canadian adolescents in different smoking stages. This study is also guided by a comprehensive theory (TTI) that targets broader and multiple influences of health-related behaviours including tobacco use.^{Footnote 21} We used a two-level multilevel logistic regression analysis to account for the clustering of students within the same school to reduce the likelihood of type 1 error.^{Footnote 48}

The study findings do not permit causal inferences due to the temporal sequence of our cross-sectional data. While self-report data are subject to response bias, the survey methodology ensured both student confidentiality and that the data were reliable and valid.Footnote 41,Footnote 42,Footnote 43 The exclusive reliance on Census data for school SES (proxy measure) has been criticized; instead the use of multiple neighbourhood measures such as physical and sociodemographic characteristics is encouraged.Footnote 44 Future research should explore TTI further by investigating the relationship between experimental smoking and other student- and school-level variables that were not available in our dataset.

Conclusion

Our findings expand on the knowledge about the student- and school-level characteristics that influence experimental smoking among secondary school students. Specifically, the characteristic of the school a student attends (i.e. being located in a rural location) can increase the likelihood of experimental smoking above and beyond individual-level influences. Our study highlights the importance of designing school-based tobacco control prevention policies and programs that are grade-sensitive and comprehensive in scope, including strategies that can increase students' attachment to their school and address multi-substance use, tobacco-related beliefs and the use of pocket money. These programs should also reach out to students who have friends and family members who smoke. Schools in rural areas may require additional resources.

Acknowledgements

The Youth Smoking Survey (YSS) is a product of the pan-Canadian capacity building project funded through a contribution agreement and contract between Health Canada and the Propel Centre for Population Health Impact from 2004 to 2011. Dr. Scott Leatherdale is a Cancer Care Ontario Research Chair in Population Studies funded by the Ontario Ministry of Health and Long-term Care. The YSS consortium includes Canadian tobacco control researchers from all provinces and provides training opportunities for university students at all levels. In Prince Edward Island, the YSS is conducted as part of the School Health Action Planning and Evaluation System – Prince Edward Island, which is funded by the Prince Edward Island Department of Education and Early Childhood Development, Prince Edward Island Department of Health and Wellness and Health Canada. Detailed information on the SHAPES/YSS-PEI system is available at: www.upei.ca/cshr.

The views expressed herein do not necessarily represent the views of Health Canada.

Conflict of interest

All authors declare that they have no conflict of interest.

References