Elsevier

Health & Place

Volume 11, Issue 1, March 2005, Pages 55-65
Health & Place

Is inter-school variation in smoking uptake and cessation due to differences in pupil composition? A cohort study

https://doi.org/10.1016/j.healthplace.2004.02.001Get rights and content

Abstract

The aims of this study were to determine if inter-school variation in smoking prevalence was due to differences in pupil composition or other school-level factors. A cohort of 13–14-year-olds (n=7147) from 52 schools was followed-up 1 year later. Random effects logistic regression was used to examine school variation in smoking uptake and cessation, with and without adjustment for pupil composition. Inter-school variation in smoking prevalence is not caused by differences in pupil composition but is due to differences in the onset of smoking arising because of unmeasured school contextual or collective factors operating on pupils’ decisions.

Introduction

There are two reasons why there may be inter-school variation in smoking incidence and prevalence. The first is that there are differences between the populations of pupils attending schools. One school, for example, may have a majority of socio-economically deprived pupils, who have parents that smoke, while another school may have a majority of pupils from affluent homes where parental smoking is uncommon. In these circumstances, it would not be surprising to find that the first school had a higher smoking prevalence than the second. Following Macintyre et al. (2002), we term these pupil-level causal factors for inter-school smoking variation compositional factors. The main explanation for a school's position in an educational attainment league table is that pupil composition varies between schools (Aveyard et al., 2004). We believe that many people assume that variation in school's smoking prevalence is due mainly to pupil composition.

A systematic review found teenage smoking was associated with 21 pupil-level factors (Tyas and Pederson, 1998). If the distribution of these pupil-level factors, i.e. differences in school composition, largely explained inter-school smoking variation then this variation would be of no epidemiological interest and research on inter-school variation could not help in designing novel teenage smoking prevention strategies. Existing school-based smoking prevention programmes have sought to modify those pupil-level risk factors for smoking that are open to modification by education in schools, which constitute a minority of risk factors for smoking. The best study methodologically was the Hutchinson Prevention Project. This intervention did not prevent smoking uptake in long-term follow-up (Peterson et al., 2000). The conclusion of the Cochrane Review on school-based programmes for smoking prevention (all based upon established pupil-level risk factors) is that there is no evidence that they are effective and that further research to produce better programmes is of debateable worth (Thomas, 2003). In such a context, the search for novel risk factors and subsequent development of novel preventative programmes might provide an alternative potentially more fruitful line of research.

The second reason why smoking prevalence might differ between schools is because some feature of the school environment influences pupils independently of pupil-level compositional risk factors for smoking. Following Macintyre et al. (2002), we term these school-level factors contextual and collective factors. The same systematic review that identified 21 ‘compositional reasons’ for variation in teenage smoking identified no school-level contextual or collective risk factors for teenage smoking (Tyas and Pederson, 1998) and there are no intervention programmes for smoking prevention that we are aware of that have only modified school-level contextual or collective factors.

Recently, we found and reviewed 22 studies describing inter-school smoking variation (Aveyard et al., 2004). We concluded that there was clear and consistent evidence of inter-school variation in pupil smoking prevalence, but the evidence that this variation was not simply a consequence of differences in pupil composition between schools was weak for two reasons.

The first reason concerns the distribution of important pupil-level risk factors for smoking. We used the systematic review by Tyas and Pederson (1998) to categorise risk factors for adolescent smoking into those that should be adjusted for because they were outside the influence of schools, and those that should be left unadjusted because they were potentially the means by which schools influence pupils’ smoking (Aveyard et al., 2004). We concluded that no studies describing inter-school variation in smoking prevalence either provided data showing the even distribution of all important pupil-level risk factors for smoking across schools, nor did they adjust for all variations in the distribution of these risk factors. Thus, pupil-level compositional differences between schools could potentially explain the observed inter-school smoking variation in many of these studies (Charlton and While, 1994; Battistich and Hom, 1997; Maes and Lievens, 1999; Wakefield et al., 2000; Moore et al., 2001).

Second, many studies were cross-sectional (e.g. Charlton and While, 1994; Battistich and Hom, 1997; Maes and Lievens, 1999; Wakefield et al., 2000; Moore et al., 2001). Implicitly, therefore, the authors were assuming that, having adjusted for a limited range of pupil-level compositional confounders, differences in smoking prevalence between schools arose because of the schools themselves. However, in cohort studies, it is invariably observed that baseline smoking status is the strongest of several other important predictors of follow-up smoking status, even when many other risk factors have been adjusted for (e.g. McNeill et al., 1988). Since baseline smoking status constitutes a strong independent risk factor for smoking at a later date, this implies that there are unknown differences between adolescents who choose to smoke and adolescents who choose not to smoke. These unknown differences cannot be adjusted out directly in cross-sectional studies. This means that cross-sectional studies that purport to explain inter-school variation by school-level factors and have adjusted for a range of pupil-level factors have not adjusted for a major pupil-level factor and this is a major threat to their validity. Cohort studies, in contrast, can partly adjust for these unknown differences between teenagers because it is possible to adjust for baseline smoking status. Claims that residual inter-school variation arises from school-level contextual and collective factors rather than pupil-level compositional factors are therefore much stronger in cohort than cross-sectional studies. In summary, the current literature provides weak evidence that compositional factors do not explain inter-school variation in smoking prevalence.

The previous review of inter-school variation showed consistent evidence that, wherever it had been sought, there was wide variation in schools’ smoking prevalence (Aveyard et al., 2004). If compositional factors predominantly explain this inter-school variation in smoking prevalence, then it may not be worth further investigation of school-level contextual or collective risk factors for smoking. If their influence were weak, teasing out their role would be a long and expensive process that would be epidemiologically challenging and then practically useless in creating novel preventative interventions because any benefit would be small. If compositional factors do not explain the wide inter-school variation, then the research necessary to understand the complex causal path between school contextual and collective factors and pupil smoking would be worthwhile because it potentially could result in large public health gains.

In the following report we describe three steps. First, we examine how much inter-school variation in smoking incidence can be explained by differences in pupil composition. Second, we assess whether inter-school smoking incidence varies because the unknown school contextual or collective factors predominantly influence whether never smokers initiate smoking or regular smokers persist with smoking or give it up. Third, we estimate the public health importance of inter-school variation implicitly caused by school-level contextual or collective factors, if it were possible to determine their influence and use such factors in a preventative programme.

Section snippets

Sampling

We used data from a previously reported trial of a school smoking prevention and cessation programme based on the Transtheoretical Model (Aveyard et al (1999), Aveyard et al (2001)). Briefly, we randomly selected 89 schools from the West Midlands, UK, and 52 (58.4%) participated in the trial. About half of the schools were in urban districts, and half in mixed urban and rural districts. Year 9 (aged 13–14) pupils (n=8352, which constituted 92.0% of all registered Year 9 pupils in these schools)

Pupil-level compositional factors

There were considerable differences in pupil composition between schools (Table 2), which might therefore be thought of as the likely explanation of much of the variation in smoking rates between schools, as is the case with respect to inter-school variation in examination success and truancy (Aveyard et al., 2004).

Quantifying inter-school smoking variation

There was statistically significant inter-school variation in regular smoking at follow-up, when differences in baseline smoking prevalence were controlled for. If a pupil attended a

Discussion

This study has shown considerable inter-school variation in smoking at follow-up, when differences in baseline smoking status of pupils between schools were accounted for. The variation could not be explained by sampling variation. Differences between schools in pupil composition actually masked some of this inter-school variation in smoking, and adjusting for inter-school variation in pupil composition only increased the unexplained differences between schools. These differences between

Conclusion

In summary, we conclude that this study provides much clearer evidence than any previous study that inter-school variation is not explained by differences in schools pupil composition. By implication, we are therefore proposing that some as yet unknown school-level contextual or collective factors have a substantial role as risk factors for the onset of smoking in adolescence. We did not aim to identify which particular contextual or collective factors responsible for this effect in this study,

Acknowledgements

The trial that the data are taken from was funded by the Health Authorities of the West Midlands. Tony Fielding has given invaluable help with the statistics and helping our understanding of multilevel modelling and we are very grateful.

References (36)

  • A. Charlton

    Why are school-based, youth-centred smoking interventions not as effective as we hoped? Some ideas for research

    International Journal of Health Promotion and Education

    (2000)
  • A. Charlton et al.

    Smoking prevalence among 16–19 year-olds related to smoking policies in sixth forms and further education

    Health Education Journal

    (1994)
  • G. Davey-Smith et al.

    Confounding in epidemiological studieswhy “independent” effects may not be all they seem

    British Medical Journal

    (1992)
  • J.R. DiFranza et al.

    Initial symptoms of nicotine dependence in adolescents

    Tobacco Control

    (2000)
  • E. Goddard

    Why Children Start Smoking. An Enquiry Carried out by Social Survey Division of OPCS on Behalf of the Department of Health

    (1990)
  • L. Jarvis

    Smoking Among Secondary School Children in 1996England

    (1997)
  • I. Kreft et al.

    Introducing Multilevel Modeling

    (1998)
  • H. Leventhal et al.

    The smoking problema review of the research and theory in behavioral risk modification

    Psychological Bulletin

    (1980)
  • Cited by (29)

    • Role theory of schools and adolescent health

      2019, The Lancet Child and Adolescent Health
    • Social capital and adolescent smoking in schools and communities: A cross-classified multilevel analysis

      2014, Social Science and Medicine
      Citation Excerpt :

      The last twenty years have witnessed an explosion of interest in place effects on health (Diez Roux, 1998; Macintyre et al., 2002). This research has found between-school and between-neighborhood variance in adolescent smoking (Aveyard et al., 2005; Henderson et al., 2008; Kelly et al., 2011; Lovato et al., 2010). However, differences in the associations of social contextual factors with the prevalence of smoking remain unclear (Galea et al., 2004).

    • Does school ethos explain the relationship between value-added education and teenage substance use? A cohort study

      2012, Social Science and Medicine
      Citation Excerpt :

      Although it is often assumed that such variation is explained by the social composition of pupils, this assumption is contradicted by the evidence. While teenage substance use varies according to characteristics such as gender or parental behaviours, these factors are unlikely to be able to account for between-school substance use variation (Aveyard, Markham, & Cheng, 2004a; Aveyard et al., 2005; West, Sweeting, & Leyland, 2004). To paraphrase Geoffrey Rose, the causes of variation in substance use between populations differ from the causes of substance use within populations (Rose, 1985).

    • Brief report: Multilevel analysis of school smoking policy and pupil smoking behaviour in Wales

      2011, Journal of Adolescence
      Citation Excerpt :

      There is mixed evidence of the effectiveness of school smoking policies in preventing adolescent smoking (Aveyard et al., 2005).

    • Adolescent smoking: Effect of school and community characteristics

      2010, American Journal of Preventive Medicine
      Citation Excerpt :

      A school effect was reported in three reviews10,22,23 of the literature on students' smoking behavior. An earlier cohort study11 found that interschool variation in smoking was not explained by differences in pupil composition but by unidentified school contextual factors. Schools as well as governments have implemented strategies aimed at lowering the prevalence of smoking in the community (e.g., educational programs, polices, and municipal regulations).

    • School differences in adolescent health and wellbeing: Findings from the Canadian Health Behaviour in School-aged Children Study

      2010, Social Science and Medicine
      Citation Excerpt :

      The findings also highlight the need to develop alternative models of data collection that both include students from the entire spectrum of the school population, and also capture changes that occur over time in a school. Cross-sectional data, as used in the HBSC, make it difficult to establish the baseline levels (Aveyard et al., 2005) that are required to make causal inferences associated with school processes. Subsequent longitudinal data collection methods will better clarify the likely complex role of the school processes we have identified.

    View all citing articles on Scopus
    View full text