Article Text

Download PDFPDF

The relationship between local clean indoor air policies and smoking behaviours in Minnesota youth
  1. E G Klein1,
  2. J L Forster2,
  3. D J Erickson2,
  4. L A Lytle2,
  5. B Schillo3
  1. 1
    Health Behavior Health Promotion Division, Ohio State University College of Public Health, Columbus, Ohio, USA
  2. 2
    Division of Epidemiology & Community Health, University of Minnesota School of Public Health, Minneapolis, Minnesota, USA
  3. 3
    ClearWay Minnesota, Minneapolis, Minnesota, USA
  1. E G Klein, Health Behavior Health Promotion Division, Ohio State University College of Public Health, 456 Cunz Hall, 1841 Neil Avenue, Columbus, OH 43210, USA; eklein{at}


Background: While clean indoor air (CIA) policies are intended to reduce exposure to second-hand smoke in the workplace, restrictions in public workplaces have the potential to discourage youth smoking. There is growing evidence from cross-sectional and ecological studies, but limited evidence from longitudinal studies that this is so.

Objective: To evaluate the association between local CIA policies and smoking behaviours among Minnesota youth over time.

Design, setting and subjects: A cohort of 4233 Minnesota youths, ages 11 to 16 at baseline, was interviewed via telephone for 6 years (2000–2006). Individual, family and community level variables were collected from participants every 6 months. A generalised estimating equation (GEE) logistic regression was used to assess the relationship between CIA policies and past-month smoking in youth over time. The analysis was controlled for potential confounders at individual and community levels.

Results: There was not significant association between CIA policies and youth smoking behaviours in the multivariate analyses. At the individual level, parental smoking significantly increased the odds of smoking nearly 40% and close friend smoking increased the odds of past-month smoking by nearly 100% for each close friend. Banning smoking in the home was significantly associated with a 12% reduction in the odds of past-month smoking.

Conclusion: After accounting for other community and individual level factors known to be associated with youth smoking, there was no significant association between CIA policies and past-month smoking for youth over time.

View Full Text

Statistics from

Improving our knowledge of modifiable factors that can influence youth smoking has great public health importance, given that an estimated 80% of smokers in the US start smoking before the age of 18.1 As a result, prevention or reduction of smoking early in life is the most efficient means of reducing the morbidity and mortality associated with smoking, and may be one of the most efficacious and appropriate public health interventions for youth.2

Youth are influenced by factors at multiple levels that encourage or discourage smoking. At the individual level, a number of modifiable interpersonal factors are known to contribute to youth smoking, including parental and close peer smoking.39 At the community level, bans on smoking in homes, restaurants and other public places, perhaps by making smoking less acceptable, have been associated with a reduction in youth smoking.1015 In particular, initial studies have suggested that smoking restrictions in public places may significantly reduce youth smoking prevalence and progression. The contribution of individual and community level factors in influencing youth smoking underscores the importance of using multilevel designs to evaluate youth smoking behaviours.16

Laws, ordinances and policies to restrict smoking in public places are generally referred to as clean indoor air (CIA) policies. Such policies are established to protect workers and others from exposure to environmental tobacco smoke (ETS),17 but may also create an environment that discourages youth smoking. Therefore, CIA policies that apply to public places may send a message to youth that smoking is not socially acceptable.18 Further, these restrictions decrease opportunities for the social exchange of cigarettes, and decrease the number of locations where youth can smoke in public, as noted by Alesci and others.10 19

Minnesota is a state in the Midwestern USA that adopted a state-level CIA policy designating smoking areas in public places in 1975, the first such policy to be established in the US. However it was not until 2000 that local cities and counties began enacting 100% smoke-free local CIA policies that applied to all worksites, and many included bars and/or restaurants. By 2007, there were 18 local CIA policies established in Minnesota; on 1 October 2007 a comprehensive CIA policy for the state was enacted.

Evidence for a relationship between CIA policies and youth smoking behaviour is limited, but growing. Much existing evidence comes from economic analyses in the US where youth cigarette consumption has declined in response to local CIA policies,20 and that CIA policies in all public places were associated with lower youth smoking prevalence.21 Further, more extensive restrictions on smoking are associated with reduced odds of youth smoking initiation and regular smoking behaviour.15 While such cross-sectional studies support the hypothesis that CIA policies are associated with a decrease in youth smoking prevalence, it is impossible to determine whether these policies caused the reduction, or whether communities with lower youth smoking are more likely to adopt a CIA policy. To date, two peer-reviewed longitudinal studies have reported a protective effect of CIA policies on youth smoking initiation and progression over time. Additional longitudinal studies are needed to confirm the strength and temporal relationship between CIA policy and youth smoking behaviour in additional cohorts. This paper assesses the influence of CIA policies on youth smoking behaviours using a multilevel analysis of longitudinal data from a cohort of youth living in Minnesota.



Data were drawn from a large, population based cohort study of Minnesota youth: the Minnesota Adolescent Community Cohort (MACC) Study. The MACC study collected data on a number of individual, family, community and state level factors that relate to youth tobacco use.

This study used a unique sampling frame that divided Minnesota into group-level geopolitical units (GPUs), boundaries thought to reflect local tobacco control environments. The state was divided into 129 GPUs, and 60 were sampled at random from 3 strata: region of the state, ethnic minority population and population density. A comparison cohort consisting of approximately 180 12–16-year-olds from metropolitan Kansas City, 180 from the rest of Kansas, 180 from North and South Dakota combined and 60 from the upper peninsula of Michigan were also recruited into the study. Given the lack of local CIA policies established in these regions, the comparison cohort was excluded from this analysis (n = 604). Based on the sampling strategy applied, these results are generalisable to adolescents in Minnesota.

A combination of probability and quota sampling methods (to assure equal age distribution) was used to establish the cohort. The goal was to recruit 60 participants, 12 each in ages 12 to 16, from each of the 60 GPUs. Recruitment was conducted by telephone by Clearwater Research (Boise, Idaho, USA), using modified random digit dial sampling. Households were called to identify those with at least one teenager between the ages of 12 and 16, and within eligible households, respondents were selected at random from among age quota cells (58.5% response rate among eligible households). Youth aged 12 to 16 were interviewed over the phone every 6 months, starting in Autumn 2000. Given that local CIA policies were not established in MACC cities in Minnesota until 2006, the period of study was determined to include participants who were exposed (to a local CIA policy) as well as unexposed. The study period ran from 2000 through 2006, which represented 11 phone surveys (surveys 1 to 12; survey 7 is missing due to a gap in funding).

The original cohort of Minnesota participants (n = 3636), as well as an additional cohort of 12-year-old participants recruited 1 year later (n = 597) were followed prospectively for 7 years, to date. Although additional 12-year-old cohorts were planned, they were not recruited due to funding limitations. A combination of incentives, multiple callbacks and extensive tracking and tracing procedures resulted in a retention rate of 77.9% at the end of 6 years.

Response rates within the MACC study average 88%, with no significant differences by gender or stage of smoking; participants who reported a non-Caucasian race were significantly less likely to complete any survey than Caucasian participants (p<0.01). This study includes data from individuals with a median of six surveys over the 6 years of the study. Participants that did not complete a specific survey were treated as missing, but were retained in the sample.


The primary outcome variable was past-month smoking, defined as self-reported response (0, 1 or more days) to the question, “Now thinking about the last 30 days, on how many of those days did you smoke a cigarette, even one or two puffs?” The primary explanatory variable of interest was CIA policy status, treated as a time-varying variable. This variable was defined as any CIA policy enacted at either a city or county level that restricted smoking beyond the state-level CIA policy in effect at the time. All of these policies completely banned smoking in restaurants; in addition, some also banned smoking in bars. We obtained data on all local Minnesota CIA policies during the study period; nine cities or counties established local CIA policies, starting in 2004. We dichotomised each CIA community in the MACC study as having either a local CIA policy that restricts smoking in restaurants and/or bars or no CIA policy where smoking areas were designated or not restricted. Each study participant was assigned CIA policy status based on city of residence at each survey, treated as a time-varying covariate.

Multilevel covariates

Variables included in the analysis were factors thought to contribute to youth smoking behaviour at either the community or individual levels. Community-level covariates were derived from US Census 2000 data,22 except for region of residence Data were assigned to study participants based on the GPU of residence at baseline. Covariates included percentage of the population aged 18 or older, percentage of the population with a college degree or higher, percentage of the region defined as rural, percentage of the population reporting Caucasian race and median household income. Region of residence was defined as rural, suburban or urban.

At the individual level, covariates included time-varying and time-invariant variables. Time-varying variables included the number of four closest friends who smoked cigarettes (0 through 4), current indoor household smoking ban (yes or no) and age (in years). Time-invariant covariates included birth cohort (age in the year 2000: 11 to 16), gender, race/ethnicity (Caucasian, African–American, Native American, Asian, Hispanic/Latino and other), maternal and paternal smoking status at baseline (yes or no).

Data analysis

Individuals within areas may be correlated, potentially due to a number of factors, including shared experiences. Given that study participants who were sampled from the same GPU may be more similar than other respondents and therefore lack independence, generalised estimating equations (GEE) logistic regression was completed, as conventional logistic regression is not appropriate for clustered data. In addition, the correlation due to repeated measures on the same individuals (non-independent observations) is taken into account by an empirical robust sandwich-type variance estimator.23 Consequently, one important advantage of GEE is that the correlation structure of the outcome need not be correctly specified as the robust estimator produces consistent point estimates and standard errors.23 Although GEE does not account for the spatial correlation of GPU as established in the MACC study design, the assessment of the intraclass correlation coefficient (ICC) of GPU was deemed ignorable (ICC = 0.01). Bivariate analyses were completed for all variables at the community and individual levels; variables that achieved a p value of 0.10 or smaller were retained in the multivariate analysis. Analyses were performed in SAS V. 9.1 (SAS, Cary, North Carolina, USA), using PROC GENMOD.24


Of the MACC cohort, 4233 participants from the state of Minnesota were included in the analysis. Given the 6 years of follow-up in this study, age, birth cohort and period effects were evaluated to determine appropriate model specification. Age was a highly statistically significant predictor of smoking behaviour, consistent with the youth smoking literature.1 Birth cohort was also significantly associated with smoking behaviours, as a younger age at baseline was a significant protective factor against smoking behaviours. An age by birth cohort interaction was evaluated and although it was statistically significant (p<0.01), visual inspection of smoking behaviours over time by birth cohort showed only one age group (baseline age of 14) had a meaningful change in slope relative to the other birth cohorts (data not shown). Due to the strong age effect and the relatively minor interaction with birth cohort shown in the unadjusted plot, a simpler model without an age by birth cohort interaction was preferred to reduce model complexity.

A description of the demographic characteristics of the sample at the community and individual level is provided in table 1, overall and stratified by CIA policy status. At the time of the most recent assessment for this study (survey 12), 1028 participants lived in an area with a local CIA policy. Overall, the participants living in an area with a CIA policy did not significantly differ in demographic characteristics from participants in areas without a CIA policy, with the exception of race/ethnicity and region of the state. The proportion of youth from each birth cohort was roughly equivalent comparing youth in the policy to those without a CIA policy. The distribution of gender was evenly balanced, with approximately 51% females in each sample. Race/ethnicity were statistically significantly different by policy status, with a higher proportion of Caucasian youth in the sample without a CIA policy (89% vs 72%) and more African–Americans in the policy sample (15% vs 2%), while other race/ethnic groups had similar proportions in both samples. Approximately 25% of both samples had a mother who was a smoker at baseline and 25% had a father who was a smoker at baseline. By the most recent survey, most participants reported a current smoking ban in the home (80% in the policy participants, 78% in the no policy participants). More of the participants in CIA policy communities lived in urban areas (47%) or suburban areas (43%); most of the participants without CIA policies lived in rural areas (64%), some suburban areas (33%) and a few urban areas (3%). At baseline, 12% of participants reported smoking in the past-month and 8% smoked in the past week; by the 5-year follow-up survey, 29% reported past-month smoking and 22% past-week smoking.

Table 1 Demographic characteristics of individuals in the Minnesota Adolescent Community Cohort (MACC) sample, by clean indoor air (CIA) policy status

At the community level (shown in table 2), the cities with a CIA policy had a higher percentage of college graduates (35% vs 23%), were significantly less rural (6% vs 43%), with a lower percentage of Caucasian residents (73% vs 93%) and a slightly higher median household income (approx. $51 000 vs approx. $48 000), compared to communities without a CIA policy, as reported in the 2000 US Census. All mean differences were statistically significant (p<0.01).

Table 2 Community-level characteristics of geopolitical units (GPUs) in the Minnesota Adolescent Community Cohort (MACC) sample, by clean indoor air (CIA) policy status

Odds ratios (ORs) for the association between community and individual covariates and past-month smoking over time are presented in table 3. The first model used unadjusted bivariate odds ratios to represent the association between individual covariates and past-month smoking over time. From the first model, covariates with a p value of 0.10 or smaller were included in the second model. The second model, referred to as a minimally adjusted model, shows odds ratios that represent the association between covariates and past-month smoking over time, adjusted for the significant community-level covariates of percentage of the population with a college education, percentage rural, percentage Caucasian, median household income and region. The third model, referred to as a fully-adjusted multivariate model, provides odds ratios that are adjusted for the same community-level variables from the second model and simultaneously adjusts for all covariates. All three models provide odds ratios and their corresponding confidence intervals for the odds of past-month smoking over time.

Table 3 Associations between characteristics at the community and individual levels with past-month smoking for adolescents over time: the Minnesota Adolescent Community Cohort (MACC)

In the unadjusted and minimally adjusted model, living in an area with a CIA policy was associated with an increased odds of past-month smoking. After adjustment for covariates at the community and individual levels in the multivariate model, youth living an area with a CIA policy had a 6% higher odds of past-month smoking, although this was not statistically significant (95% CI 0.93 to 1.21, p = 0.41).

In the fully adjusted multivariate model, age produced a consistent, positive effect on smoking behaviours; every 1-year increase in age was associated with a 26% increase in odds of past-month smoking. (Gender was the only individual-level variable that was not statistically significant in the unadjusted bivariate model, therefore it was not retained in the adjusted models.) The odds of smoking behaviours varied by racial and ethnic group, with African–Americans, Asians and Hispanic/Latinos at a lower odds of past-month smoking compared to Caucasians; Native Americans and other races at a consistently higher odds past-month smoking, compared to Caucasians. For every close friend who smoked, there was a 96% increased odds in past-month smoking, compared to having no close friends who smoke (95% CI 1.88 to 2.03). Parental smoking significantly increased the odds of past-month smoking by nearly 40% (maternal smoking OR 1.38, p<0.01; paternal smoking OR 1.39, p<0.01). Living in a home with a ban on smoking for residents and guests had a consistent, protective effect that reduced past-month smoking by 12% after multivariate adjustment (OR 0.88, 95% CI 0.80 to 0.96).


CIA policies are known to protect workers and the public from exposure to ETS, and are deemed the most effective means to do so.17 Cross-sectional findings at the local and state level have demonstrated a reduction in youth smoking behaviours associated with local CIA policies.15 25 26 Siegel and colleagues that suggest that local CIA policies may result in a reduction in youth smoking behaviours over time.27 28 As noted by Wakefield and Forster, the use of a prospective cohort study provides temporal evidence for the causal association between CIA policies and youth smoking behaviours.29

What this paper adds

  • This evaluation found no significant association between the enactment of local CIA policies and a reduction in youth smoking behaviours.

  • This multilevel analysis accounts for individual and community level factors associated with youth smoking, and the prospective design provides additional assurance of the temporal association.

Yet, our findings are not consistent with the strength or direction of the previous studies. Siegel et al found between a 40% decreased odds and a 61% decreased adjusted risk of progression to regular smoking over time.27 28 Our findings suggested that youth living in an area with a CIA policy were at 6% increased risk of past-month smoking, although these results were not statistically significant. Our primary explanation for this inconsistency is the influence of more proximal factors. Close friend smoking, parental smoking and home smoking bans produced a substantial, significant influence on smoking behaviours that resulted in mitigation of the magnitude of the association between CIA policy and youth smoking.

The period of late adolescence to young adulthood may be a time when smoking patterns are in flux,3032 making a cohesive statement more challenging. The youth in the MACC sample were aged 16 or older at the time the first local CIA policy was established. Therefore our study may have missed the most critical period of the development of smoking behaviours between ages 10 and 15,25 resulting in a bias toward the null hypothesis. While late adolescence to early adulthood remains a critical period of development of smoking behaviour, more information is needed regarding the influences at individual and community levels on youth during this time frame.

As noted by Albers et al, the perceived prevalence of smoking in the community influences youth smoking behaviours, and we would expect to see a reduction in youth smoking in policy communities compared to communities without CIA policies.18 In our study, the influence of friend smoking status rendered a more powerful influence on smoking behaviour than CIA policy: youth with close friends who smoked were more likely to smoke than youth with no close friends who smoke, which is consistent with other studies.3 7 8 As noted by Forster et al, youth access cigarettes primarily through social sources,19 so it may be reasonable to assume that youth with friends who smoke are the most likely to be influenced by environmental barriers like CIA policies as such policies reduce opportunities to smoke outside the house and/or the social exchange of cigarettes, but this is not what we found.

Given the multilevel structure to this research question, there is potential for covariates to mediate the association between CIA policy and youth smoking. Specifically, parental smoking and home smoking policies may be influenced by the establishment of CIA policies at the community level.33 34 Close friend smoking, especially during later adolescence,8 may also serve as a mediating variable between CIA policy and youth smoking. According to MacKinnon et al, mediation and confounding are statistically equivalent,35 so the multilevel modelling strategy to address this question is assumed to be appropriate, regardless of the nature of the relationship between CIA policies and these variables. Lastly, while residual confounding is possible between age, parent smoking, or close friend smoking and past-month smoking, we believe that residual confounding would be minimal given the strong associations described in this analysis.

At the community level, the majority of CIA policies in Minnesota were established in urban or suburban areas. This concentration of policies in suburban and urban areas makes the exploration of regional differences in smoking very difficult. Without the ability to randomise communities to policy status, research relies on observational data that may not be well balanced across all relevant characteristics. Additional research is needed into the role of regional variation in the adoption of local CIA policies. With more than 500 local communities in the US with established CIA policies for 100% smoke-free workplaces and many more worldwide,36 replication in different regions of the US and other countries could be completed to help explore the potential for regional differences.

In addition, the lack of random assignment to policy status could have allowed for unidentified factors to confound the association between CIA policies and youth smoking. While most of the major individual-level smoking predictors have been included in the analytic models, a number of community-level factors that are known to be associated with youth smoking prevalence, including cigarette pricing, marketing of tobacco products, anti-tobacco media campaigns and other tobacco control efforts were not measured or did not vary within this study.3739 It is possible that the observed effects of CIA policies could be attributed to these or other unmeasured confounders, rather than to CIA policies.

The level of complexity of the data and analytic method meant that we were unable to test the interaction between age, birth cohort and period effects. However, the inclusion of age and birth cohort effects very likely account for the strongest trends to be observed in these data. Additionally, future analysis of adolescent or young adult cohort studies should explore the role of friend of smoking as an intermediate variable in the relationship between CIA policy and youth smoking.

This study suggests that CIA policies in Minnesota cities did not significantly influence youth smoking behaviours. These direction of these findings are inconsistent with what has been suggested from cross-sectional data and limited longitudinal studies—youth living in areas where smoking is restricted in workplaces may be less likely to smoke. Clearly, more research is needed to clarify the role of CIA policies and their association with youth smoking. Despite these findings, it is clear that CIA policies are an effective intervention to protect employees and the public from exposure to ETS.


View Abstract


  • Competing interests: None.

  • Funding: This work was supported by grants CA86191 from the National Institutes of Health and RC 2006-0047 from ClearWay Minnesota. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of ClearWay Minnesota.

  • Ethics approval: The University of Minnesota Institutional Review Board approved this study.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.