Downloaded from http://tobaccocontrol.bmj.com/ on January 1, 2016 - Published by group.bmj.com
TC Online First, published on October 30, 2015 as 10.1136/tobaccocontrol-2015-052531 Research paper
School personnel smoking, school-level policies, and adolescent smoking in low- and middle-income countries Silda Nikaj,1 Frank Chaloupka2 ▸ Additional material is published online only. To view please visit the journal online (http://dx.doi.org/10.1136/ tobaccocontrol-2015-052531). 1
Department of Economics, Texas Christian University, Fort Worth, Texas, USA 2 Department of Economics, Institute for Health Research and Policy, University of Illinois at Chicago, Chicago, Illinois, USA Correspondence to Dr Silda Nikaj, Department of Economics, Texas Christian University, TCU Box 298510, 2800 S. University Drive, Fort Worth, TX 76129, USA; [email protected] Received 15 June 2015 Accepted 9 October 2015
ABSTRACT Objectives This paper examines the link between personnel and teacher smoking on school grounds, and student smoking in 62 low-income and middle-income countries. Methods We use a two-part model to estimate the effect of smoking by school personnel on youth smoking. In the first part, we model the decision to smoke for all students, using a linear probability model. In the second part, we estimate cigarette consumption among smokers. We employ country fixed effects to address country-level time-invariant unobservable factors and control for an array of local-level variables to address local-level heterogeneity. Results We find that smoking by personnel and teachers on school grounds is associated with higher smoking prevalence among all youths, and higher cigarette consumption among female smokers. Our findings suggest that consumption among female smokers is primarily affected by smoking among female personnel, and that younger personnel/teachers appear to be more influential in determining behaviours among young people. In addition, we find that smoking restrictions on staff are associated with reductions in average consumption among female students. Conclusions Low-income and middle-income countries may reduce smoking among young people by banning smoking for teachers and school personnel on school grounds.
INTRODUCTION
To cite: Nikaj S, Chaloupka F. Tob Control Published Online First: [please include Day Month Year] doi:10.1136/ tobaccocontrol-2015052531
Because of the burden of death and disease caused by smoking, developed countries have invested in comprehensive tobacco control programmes that have curbed smoking among population groups.1 2 Smoking presents a growing public health problem in low-income and middle-income countries where lack of smoking restrictions, low levels of cigarette taxation, and high affordability of cigarettes have contributed to increasing levels of cigarette use.1–6 Recent research on low-income and middle-income countries finds that young people are more responsive to cigarette price changes than youth in developed countries, highlighting the importance of taxation as a policy tool in reducing smoking among the populations of developing countries.7–15 In recent years, the impact of social factors on health behaviours has received increasing attention from researchers. This research arises from the understanding that people make choices within a social context, and that the social context may shift the costs and benefits of particular actions.16–18 An
extensive amount of literature in developed countries finds that peer and parental smoking have large impacts on adolescent smoking.19–25 Similarly, evidence suggests that school environments and restrictions on smoking in school may reduce smoking prevalence and consumption among youth.26–33 Despite the extensive evidence linking social environments and youth smoking in high-income countries, little evidence exists on the impact of school environments and smoking among youth in lowincome and middle-income countries.31 This paper is the first to investigate associations among school personnel smoking, school-level policies and adolescent smoking behaviours, in 62 low-income and middle-income countries. School environments and school personnel can affect behaviours among young people along several pathways. First, behaviour of school personnel likely serves as a model for appropriate behaviour for young people. If smoking among school personnel is common, then young people are likely to interpret such signals as appropriate or ‘adult’ behaviour.18 Second, it becomes increasingly difficult for personnel who smoke to discuss the dangers of smoking with students. Third, personnel who smoke on school grounds may be less likely to enforce smoking restrictions among students. Finally, personnel who smoke may stimulate consumption among student smokers not only by exposing them to secondhand smoke but also providing the visual stimuli or cues for smoking on school grounds.34 Researchers face several challenges in estimating the impact of personnel smoking on youth smoking. Unobserved factors and school selection/ sorting confound the effect of social influence on individual behaviours.35 Personnel and students may select schools based on smoking behaviours or other unobserved characteristics correlated with smoking. Simple associations between personnel and youth smoking would capture both the effect of personnel smoking behaviours and sorting. Without appropriate controls, sorting would overstate the true impact of school personnel’s behaviours on youth smoking. Second, bi-directionality between individual behaviour and the behaviours of those in one’s social circle presents further methodological challenges. In the peer context, it is unclear whether peer smoking is affecting individual smoking as there is invariably some effect exerted from individual smoking on peer smoking. The bi-directionality is especially pronounced among individuals who occupy the same social standing in a network, but
Nikaj S, Chaloupka F. Tob Control 2015;0:1–7. doi:10.1136/tobaccocontrol-2015-052531
Copyright Article author (or their employer) 2015. Produced by BMJ Publishing Group Ltd under licence.
1
Downloaded from http://tobaccocontrol.bmj.com/ on January 1, 2016 - Published by group.bmj.com
Research paper researchers see it as less of a concern among individuals of different social standings.36 This line of reasoning posits that school personnel occupy a position of authority and exhibit some social distance from students, and thus are unlikely to be affected by student smoking behaviour directly. Finally, high rates of smoking among students and personnel may be the result of an environment that lacks tobacco control restrictions. Naïve analyses would produce high rates of correlation in smoking behaviours among population groups; however, smoking among all groups would be influenced by shared contextual factors that affect everyone in a social network. Without controls for sorting of students and personnel into schools, the bi-directionality of influences in a group, and the local-level environment, estimates of the behavioural response to smoking in one’s social network will be biased. This paper estimates the associations between smoking by school personnel, school-level restrictions, and smoking by middle and high school students for 62 countries that participated in the Global Youth Tobacco Survey (GYTS) and Global School Personnel Survey (GSPS). We hypothesise that students exposed to teachers and personnel who smoke on school grounds, will be more likely to smoke and have higher levels of average cigarette consumption. Furthermore, we test the hypothesis that stronger smoking restrictions on school grounds are associated with lower levels of smoking prevalence and consumption among young people. We control for an array of locallevel, tobacco-related characteristics that reduce the likelihood of bias in estimates,i and capture country-specific characteristics that are unobservable, through the use of country fixed effects.ii
DATA AND METHODS The data used are pooled cross-sections from 62 countries of GYTS and GSPS from 2002 to 2008. The GYTS is a schoolbased survey that examines cigarette use, knowledge and attitudes among youths 11–19 years of age. The survey uses a sample design that chooses schools at random within a given country and then randomly surveys classes within these schools. Within the same schools, a second survey (GSPS) examined cigarette use, knowledge and attitudes about smoking among school personnel. The survey was voluntary and not all schools participating in the GYTS participated in the GSPS, thus limiting the sample of countries that participated in both surveys to a total of 77. Many of the survey questions vary over time, which does not allow exact matching of control variables across different waves of the survey for several countries. The survey question mismatch and exclusion of high-income countries limits the number of countries to a total of 62 for the current analysis. A detailed list of excluded countries as well as those in the analysis is provided in online supplementary appendix A. We employ two measures of smoking in our analysis. Smoking participation is a dichotomous variable that takes a value of 1 if the student smoked at least one cigarette in the
i
Our data do not allow for the use of school fixed effects, or school level dummy variables. A school fixed effects methodology, the standard method to deal with selection when data structures allow, would require some form of variation in personnel smoking within schools (or over time) in order to identify the impact of smoking at school by school staff. Given the cross-sectional structure of the data there is no way to link schools over time. Furthermore, the data do not allow us to link average personnel smoking to individual classes within the school. ii Each country has its own dummy or indicator variable in order to account for country-level unobservable factors that affect smoking and are correlated with our other control variables. 2
month before the survey, and 0 otherwise. We construct smoking intensity (consumption) by multiplying the number of days that smoking occurred in the past month by the average number of cigarettes smoked daily. Table 1 provides summary statistics of the analysis variables. The all-sample smoking prevalence is 11%. The average smoker consumes 53 cigarettes a month. Individual-level explanatory variables include age, gender, measures of parental smoking and whether the youth had instruction in school about the dangers of smoking in the last year. The variable of interest for this analysis is the share of school personnel (personnel smoking) who smoke on school grounds. We constructed the variable by aggregating school personnel responses from the GSPS to a question of whether they had ever smoked cigarettes on school property during the last year. On average, 13% of all personnel had smoked on school property in the past year, while close to 18% of all school personnel are smokers. Over 70% of smokers among personnel continue smoking on school premises, suggesting that smoking is commonplace in schools. We include several variables that control for the local tobacco-related environment: exposure to antismoking media, exposure to cigarette advertising and access to commercial cigarettes. We constructed all variables by aggregating student responses at the school level in order to reduce potential endogeneityiii of individual responses. Exposure to antismoking media is the percentage of non-smoking students who report recent exposure to antismoking media messages in broadcast and print media. We define exposure to cigarette advertising as the percentage of non-smoking students who report recent exposure to advertising in newspapers and billboards.iii Reduced access to commercial cigarettes is the percentage of students who report having been denied cigarette sales by local vendors due to their age. Finally, we include controls for smoking restrictions on school grounds. More particularly, we included three variables by aggregating responses from the GSPS survey at the school level. The survey asked personnel whether the school had a ban on smoking on school grounds for students (student ban), a ban for personnel (staff ban), and whether the school enforced the ban for personnel as well as for students (enforcement). These variables do not capture the impact of smoke-free policies, as from the data it is impossible to discern which schools have instituted smoke-free policies, but they may provide an estimate of the plausible impact of such policies.iv
iii Individual-level responses to exposure variables (antismoking media and cigarette advertising) suffer from potential reverse causality, since smokers are likely targeted by both cigarette advertising and counter-advertising, and have a higher propensity to notice cigarette advertising than non-smokers.37 38 If we included smokers in generating exposure variables, these variables would represent both the effect of ‘targeting’ and the effect these policies have on smoking. Thus we define both variables among non-smokers in order to identify the effect of exposure and not that of targeting/selection. iv We are aware of the concern in generating these variables from the survey data, as they likely also reflect knowledge/awareness of such bans in addition to the actual presence of a ban. However, we believe that leaving these variables out would produce bias in our personnel smoking variable. The bias would arise because we would not be able to separate the effect of personnel smoking, per se, from enforcement of policies by personnel who smoke. Leaving enforcement variables out would likely bias our estimates of personnel smoking. We therefore include these variables but caution that the estimates ought to be interpreted carefully and do not represent actual policy variables.
Nikaj S, Chaloupka F. Tob Control 2015;0:1–7. doi:10.1136/tobaccocontrol-2015-052531
Downloaded from http://tobaccocontrol.bmj.com/ on January 1, 2016 - Published by group.bmj.com
Research paper Table 1 Definition of analysis variables and summary statistics Variables
Definitions
Full sample
Males
Females
Smoking participation Average consumption Age Male Parental smoking Father Mother Both Class on harms of smoking Exposure to cigarette advertising Exposure to antismoking media
1=Reported smoking cigarettes in the last 30 days, 0=otherwise Number of cigarettes smoked in the past month, conditional on being a smoker Respondent’s age in years 1=male, 0=female 1 if parents smoke, 0 otherwise
0.11 (0.31) 53.19 (120.76) 14.36 (1.63) 0.51 (0.50)
0.15 (0.36) 56.12 (122.88)
0.07 (0.25) 46.37 (115.42)
Reduced access to commercial cigarettes Personnel smoking Teachers smoking Student smoking ban Staff smoking ban Policy enforcement
1=if student had a class on harms of smoking in past year, 0 otherwise Proportion of non-smoking survey participants who report recent exposure to cigarette advertising in print media Proportion of non-smoking survey participants who report recent exposure to antismoking media messages Proportion of survey participants who report being denied cigarette sales due to age in the month prior to the survey Proportion of school personnel smoking at school Proportion of teachers smoking at school Proportion of school personnel who report a student smoking ban exists Proportion of school personnel who report a staff smoking ban exists Proportion of staff who report ban is enforced on students and staff
0.29 0.02 0.05 0.64 0.45
(0.45) (0.15) (0.22) (0.53) (0.14)
0.73 (0.19) 0.41 (0.25) 0.13 0.13 0.79 0.54 0.79
(0.16) (0.16) (0.22) (0.27) (0.27)
SDs in parentheses.
To control for country-specific unobservable characteristics that could drive smoking, we include country fixed effects or dummy variables. We capture changes in smoking over time by including linear and quadratic time trends. Because GYTS and GSPS survey most countries only once, year dummies would be almost perfectly collinear with country dummies and cannot be utilised. To address missing responses, we impute missing data for control variables.39
Empirical specification We use a two-part model to estimate the effect of smoking by school personnel and school-level variables on youth smoking. In the first part, we model smoking participation for all students using a linear probability model. In the second part, we estimate cigarette consumption among smokers. The conditional demand estimation uses a generalised linear model (GLM) with a Gamma distribution and log-link.40 41 We cluster all standard errors at the school level.
RESULTS OLS results for smoking participation Table 2 summarises results for smoking prevalence. We only present the fully specified model. We ran several models (not shown) by including/excluding different controls, to make sure that our estimates were robust and did not change. The point estimate—0.063—of personnel smoking at school is highly significant. A 10-percentage point increase in smoking on school grounds by personnel is associated with an increase in smoking prevalence of 0.63 percentage points—a 5.7% increase in smoking prevalence. Smoking by school personnel is linked to a larger absolute effect on smoking prevalence among boys than among girls ( parameter estimates of 0.069 vs 0.05). However, the relative effects on smoking prevalence for a 10-percentage point increase in smoking among personnel are 4.6% and 7.1% for boys and girls, respectively. Nikaj S, Chaloupka F. Tob Control 2015;0:1–7. doi:10.1136/tobaccocontrol-2015-052531
Other controls have the expected signs. Males are more likely to smoke than females. Maternal smoking and paternal smoking are associated with increases in the probability of smoking by 11.1 and 4.6 percentage points, respectively, highlighting the large influence parents can exert on their children’s smoking. Greater exposure to cigarette advertising is associated with increased smoking prevalence. Exposure to antismoking media is linked with a decrease in prevalence, but the association is significant only at the 10% level. Being denied sale of cigarettes due to one’s age is associated with a reduction in smoking participation among girls. Instruction at school about the dangers of smoking is linked to a lower probability of smoking for all youths. On smoking restrictions at school, the parameter on having a student ban is insignificant or has a positive association (for females). The most likely interpretation is that schools with smoking bans often have a larger share of student smokers. In fact, schools often adopt such policies if smoking among the student body becomes a persistent or increasing problem, leading to a positive correlation between smoking bans and smoking among students. A second potential explanation may be that having a ban may not reduce participation rates unless schools enforce such bans comprehensively. The inclusion of an interaction between the enforcement variable and the student ban variable suggests the effect is zero. This finding should not be surprising, as our variables are likely capturing the effect of both policy selection—where schools with a large share of smokers are more likely to adopt smoking bans among students —and the effect of the ban. Since policy selection is positively associated with smoking participation, and bans and enforcement of such bans are negatively associated with smoking, the net effect is comprised of these two opposing effects, which could explain our null findings. A ban on smoking for personnel is associated with reductions in the probability that girls smoke (significant at the 10% level). Adding the interaction of enforcement and smoking bans for personnel does not affect the decision to smoke among youth. 3
Downloaded from http://tobaccocontrol.bmj.com/ on January 1, 2016 - Published by group.bmj.com
Research paper Table 2
Linear probability model of smoking participation
Staff smoking Student ban Staff ban
All†
Males†
Females†
All†
Males†
Females†
0.063*** (0.013) 0.019+ (0.011) −0.014 (0.009)
0.070*** (0.017) 0.016 (0.016) −0.011 (0.012)
0.050*** (0.012) 0.022* (0.010) −0.013+ (0.008)
−0.149*** (0.011) 0.006*** (0.000) 0.081*** (0.002) 0.111*** (0.008) 0.046*** (0.002) 0.120*** (0.006) 0.098*** (0.021) −0.012+ (0.006) −0.034 (0.021) −0.010*** (0.002) 229 995
−0.171*** (0.016) 0.007*** (0.001)
−0.120*** (0.013) 0.004*** (0.000)
0.069*** (0.017) 0.01 (0.022) 0.004 (0.027) 0.009 (0.024) −0.02 (0.033) −0.171*** (0.016) 0.007*** (0.001)
0.050*** (0.012) 0.02 (0.013) −0.015 (0.018) 0.002 (0.015) 0.002 (0.022) −0.120*** (0.013) 0.004*** (0.000)
0.116*** (0.012) 0.064*** (0.003) 0.138*** (0.009) 0.060* (0.031) −0.014 (0.010) −0.022 (0.033) −0.009*** (0.003) 115 284
0.105*** (0.008) 0.030*** (0.002) 0.103*** (0.007) 0.127*** (0.018) −0.008 (0.005) −0.037* (0.017) −0.009*** (0.002) 114 711
0.063*** (0.013) 0.021 (0.015) −0.01 (0.020) −0.001 (0.017) −0.005 (0.024) −0.149*** (0.011) 0.006*** (0.000) 0.081*** (0.002) 0.111*** (0.008) 0.046*** (0.002) 0.120*** (0.006) 0.098*** (0.021) −0.012+ (0.006) −0.034 (0.021) −0.010*** (0.002) 229 995
0.116*** (0.012) 0.064*** (0.003) 0.138*** (0.009) 0.060* (0.031) −0.014 (0.010) −0.022 (0.033) −0.009*** (0.003) 115 284
0.105*** (0.008) 0.030*** (0.002) 0.103*** (0.007) 0.127*** (0.018) −0.008 (0.005) −0.037* (0.017) −0.009*** (0.002) 114 711
Student ban enforcement Staff ban enforcement Age Age squared Male Mother smokes Father smokes Both parents smoke Exposure to cigarette ads Exposure to antismoking ads Reduced access Class on smoking N
All specifications include country fixed effects, and linear and quadratic time trends. +p
TC Online First, published on October 30, 2015 as 10.1136/tobaccocontrol-2015-052531 Research paper
School personnel smoking, school-level policies, and adolescent smoking in low- and middle-income countries Silda Nikaj,1 Frank Chaloupka2 ▸ Additional material is published online only. To view please visit the journal online (http://dx.doi.org/10.1136/ tobaccocontrol-2015-052531). 1
Department of Economics, Texas Christian University, Fort Worth, Texas, USA 2 Department of Economics, Institute for Health Research and Policy, University of Illinois at Chicago, Chicago, Illinois, USA Correspondence to Dr Silda Nikaj, Department of Economics, Texas Christian University, TCU Box 298510, 2800 S. University Drive, Fort Worth, TX 76129, USA; [email protected] Received 15 June 2015 Accepted 9 October 2015
ABSTRACT Objectives This paper examines the link between personnel and teacher smoking on school grounds, and student smoking in 62 low-income and middle-income countries. Methods We use a two-part model to estimate the effect of smoking by school personnel on youth smoking. In the first part, we model the decision to smoke for all students, using a linear probability model. In the second part, we estimate cigarette consumption among smokers. We employ country fixed effects to address country-level time-invariant unobservable factors and control for an array of local-level variables to address local-level heterogeneity. Results We find that smoking by personnel and teachers on school grounds is associated with higher smoking prevalence among all youths, and higher cigarette consumption among female smokers. Our findings suggest that consumption among female smokers is primarily affected by smoking among female personnel, and that younger personnel/teachers appear to be more influential in determining behaviours among young people. In addition, we find that smoking restrictions on staff are associated with reductions in average consumption among female students. Conclusions Low-income and middle-income countries may reduce smoking among young people by banning smoking for teachers and school personnel on school grounds.
INTRODUCTION
To cite: Nikaj S, Chaloupka F. Tob Control Published Online First: [please include Day Month Year] doi:10.1136/ tobaccocontrol-2015052531
Because of the burden of death and disease caused by smoking, developed countries have invested in comprehensive tobacco control programmes that have curbed smoking among population groups.1 2 Smoking presents a growing public health problem in low-income and middle-income countries where lack of smoking restrictions, low levels of cigarette taxation, and high affordability of cigarettes have contributed to increasing levels of cigarette use.1–6 Recent research on low-income and middle-income countries finds that young people are more responsive to cigarette price changes than youth in developed countries, highlighting the importance of taxation as a policy tool in reducing smoking among the populations of developing countries.7–15 In recent years, the impact of social factors on health behaviours has received increasing attention from researchers. This research arises from the understanding that people make choices within a social context, and that the social context may shift the costs and benefits of particular actions.16–18 An
extensive amount of literature in developed countries finds that peer and parental smoking have large impacts on adolescent smoking.19–25 Similarly, evidence suggests that school environments and restrictions on smoking in school may reduce smoking prevalence and consumption among youth.26–33 Despite the extensive evidence linking social environments and youth smoking in high-income countries, little evidence exists on the impact of school environments and smoking among youth in lowincome and middle-income countries.31 This paper is the first to investigate associations among school personnel smoking, school-level policies and adolescent smoking behaviours, in 62 low-income and middle-income countries. School environments and school personnel can affect behaviours among young people along several pathways. First, behaviour of school personnel likely serves as a model for appropriate behaviour for young people. If smoking among school personnel is common, then young people are likely to interpret such signals as appropriate or ‘adult’ behaviour.18 Second, it becomes increasingly difficult for personnel who smoke to discuss the dangers of smoking with students. Third, personnel who smoke on school grounds may be less likely to enforce smoking restrictions among students. Finally, personnel who smoke may stimulate consumption among student smokers not only by exposing them to secondhand smoke but also providing the visual stimuli or cues for smoking on school grounds.34 Researchers face several challenges in estimating the impact of personnel smoking on youth smoking. Unobserved factors and school selection/ sorting confound the effect of social influence on individual behaviours.35 Personnel and students may select schools based on smoking behaviours or other unobserved characteristics correlated with smoking. Simple associations between personnel and youth smoking would capture both the effect of personnel smoking behaviours and sorting. Without appropriate controls, sorting would overstate the true impact of school personnel’s behaviours on youth smoking. Second, bi-directionality between individual behaviour and the behaviours of those in one’s social circle presents further methodological challenges. In the peer context, it is unclear whether peer smoking is affecting individual smoking as there is invariably some effect exerted from individual smoking on peer smoking. The bi-directionality is especially pronounced among individuals who occupy the same social standing in a network, but
Nikaj S, Chaloupka F. Tob Control 2015;0:1–7. doi:10.1136/tobaccocontrol-2015-052531
Copyright Article author (or their employer) 2015. Produced by BMJ Publishing Group Ltd under licence.
1
Downloaded from http://tobaccocontrol.bmj.com/ on January 1, 2016 - Published by group.bmj.com
Research paper researchers see it as less of a concern among individuals of different social standings.36 This line of reasoning posits that school personnel occupy a position of authority and exhibit some social distance from students, and thus are unlikely to be affected by student smoking behaviour directly. Finally, high rates of smoking among students and personnel may be the result of an environment that lacks tobacco control restrictions. Naïve analyses would produce high rates of correlation in smoking behaviours among population groups; however, smoking among all groups would be influenced by shared contextual factors that affect everyone in a social network. Without controls for sorting of students and personnel into schools, the bi-directionality of influences in a group, and the local-level environment, estimates of the behavioural response to smoking in one’s social network will be biased. This paper estimates the associations between smoking by school personnel, school-level restrictions, and smoking by middle and high school students for 62 countries that participated in the Global Youth Tobacco Survey (GYTS) and Global School Personnel Survey (GSPS). We hypothesise that students exposed to teachers and personnel who smoke on school grounds, will be more likely to smoke and have higher levels of average cigarette consumption. Furthermore, we test the hypothesis that stronger smoking restrictions on school grounds are associated with lower levels of smoking prevalence and consumption among young people. We control for an array of locallevel, tobacco-related characteristics that reduce the likelihood of bias in estimates,i and capture country-specific characteristics that are unobservable, through the use of country fixed effects.ii
DATA AND METHODS The data used are pooled cross-sections from 62 countries of GYTS and GSPS from 2002 to 2008. The GYTS is a schoolbased survey that examines cigarette use, knowledge and attitudes among youths 11–19 years of age. The survey uses a sample design that chooses schools at random within a given country and then randomly surveys classes within these schools. Within the same schools, a second survey (GSPS) examined cigarette use, knowledge and attitudes about smoking among school personnel. The survey was voluntary and not all schools participating in the GYTS participated in the GSPS, thus limiting the sample of countries that participated in both surveys to a total of 77. Many of the survey questions vary over time, which does not allow exact matching of control variables across different waves of the survey for several countries. The survey question mismatch and exclusion of high-income countries limits the number of countries to a total of 62 for the current analysis. A detailed list of excluded countries as well as those in the analysis is provided in online supplementary appendix A. We employ two measures of smoking in our analysis. Smoking participation is a dichotomous variable that takes a value of 1 if the student smoked at least one cigarette in the
i
Our data do not allow for the use of school fixed effects, or school level dummy variables. A school fixed effects methodology, the standard method to deal with selection when data structures allow, would require some form of variation in personnel smoking within schools (or over time) in order to identify the impact of smoking at school by school staff. Given the cross-sectional structure of the data there is no way to link schools over time. Furthermore, the data do not allow us to link average personnel smoking to individual classes within the school. ii Each country has its own dummy or indicator variable in order to account for country-level unobservable factors that affect smoking and are correlated with our other control variables. 2
month before the survey, and 0 otherwise. We construct smoking intensity (consumption) by multiplying the number of days that smoking occurred in the past month by the average number of cigarettes smoked daily. Table 1 provides summary statistics of the analysis variables. The all-sample smoking prevalence is 11%. The average smoker consumes 53 cigarettes a month. Individual-level explanatory variables include age, gender, measures of parental smoking and whether the youth had instruction in school about the dangers of smoking in the last year. The variable of interest for this analysis is the share of school personnel (personnel smoking) who smoke on school grounds. We constructed the variable by aggregating school personnel responses from the GSPS to a question of whether they had ever smoked cigarettes on school property during the last year. On average, 13% of all personnel had smoked on school property in the past year, while close to 18% of all school personnel are smokers. Over 70% of smokers among personnel continue smoking on school premises, suggesting that smoking is commonplace in schools. We include several variables that control for the local tobacco-related environment: exposure to antismoking media, exposure to cigarette advertising and access to commercial cigarettes. We constructed all variables by aggregating student responses at the school level in order to reduce potential endogeneityiii of individual responses. Exposure to antismoking media is the percentage of non-smoking students who report recent exposure to antismoking media messages in broadcast and print media. We define exposure to cigarette advertising as the percentage of non-smoking students who report recent exposure to advertising in newspapers and billboards.iii Reduced access to commercial cigarettes is the percentage of students who report having been denied cigarette sales by local vendors due to their age. Finally, we include controls for smoking restrictions on school grounds. More particularly, we included three variables by aggregating responses from the GSPS survey at the school level. The survey asked personnel whether the school had a ban on smoking on school grounds for students (student ban), a ban for personnel (staff ban), and whether the school enforced the ban for personnel as well as for students (enforcement). These variables do not capture the impact of smoke-free policies, as from the data it is impossible to discern which schools have instituted smoke-free policies, but they may provide an estimate of the plausible impact of such policies.iv
iii Individual-level responses to exposure variables (antismoking media and cigarette advertising) suffer from potential reverse causality, since smokers are likely targeted by both cigarette advertising and counter-advertising, and have a higher propensity to notice cigarette advertising than non-smokers.37 38 If we included smokers in generating exposure variables, these variables would represent both the effect of ‘targeting’ and the effect these policies have on smoking. Thus we define both variables among non-smokers in order to identify the effect of exposure and not that of targeting/selection. iv We are aware of the concern in generating these variables from the survey data, as they likely also reflect knowledge/awareness of such bans in addition to the actual presence of a ban. However, we believe that leaving these variables out would produce bias in our personnel smoking variable. The bias would arise because we would not be able to separate the effect of personnel smoking, per se, from enforcement of policies by personnel who smoke. Leaving enforcement variables out would likely bias our estimates of personnel smoking. We therefore include these variables but caution that the estimates ought to be interpreted carefully and do not represent actual policy variables.
Nikaj S, Chaloupka F. Tob Control 2015;0:1–7. doi:10.1136/tobaccocontrol-2015-052531
Downloaded from http://tobaccocontrol.bmj.com/ on January 1, 2016 - Published by group.bmj.com
Research paper Table 1 Definition of analysis variables and summary statistics Variables
Definitions
Full sample
Males
Females
Smoking participation Average consumption Age Male Parental smoking Father Mother Both Class on harms of smoking Exposure to cigarette advertising Exposure to antismoking media
1=Reported smoking cigarettes in the last 30 days, 0=otherwise Number of cigarettes smoked in the past month, conditional on being a smoker Respondent’s age in years 1=male, 0=female 1 if parents smoke, 0 otherwise
0.11 (0.31) 53.19 (120.76) 14.36 (1.63) 0.51 (0.50)
0.15 (0.36) 56.12 (122.88)
0.07 (0.25) 46.37 (115.42)
Reduced access to commercial cigarettes Personnel smoking Teachers smoking Student smoking ban Staff smoking ban Policy enforcement
1=if student had a class on harms of smoking in past year, 0 otherwise Proportion of non-smoking survey participants who report recent exposure to cigarette advertising in print media Proportion of non-smoking survey participants who report recent exposure to antismoking media messages Proportion of survey participants who report being denied cigarette sales due to age in the month prior to the survey Proportion of school personnel smoking at school Proportion of teachers smoking at school Proportion of school personnel who report a student smoking ban exists Proportion of school personnel who report a staff smoking ban exists Proportion of staff who report ban is enforced on students and staff
0.29 0.02 0.05 0.64 0.45
(0.45) (0.15) (0.22) (0.53) (0.14)
0.73 (0.19) 0.41 (0.25) 0.13 0.13 0.79 0.54 0.79
(0.16) (0.16) (0.22) (0.27) (0.27)
SDs in parentheses.
To control for country-specific unobservable characteristics that could drive smoking, we include country fixed effects or dummy variables. We capture changes in smoking over time by including linear and quadratic time trends. Because GYTS and GSPS survey most countries only once, year dummies would be almost perfectly collinear with country dummies and cannot be utilised. To address missing responses, we impute missing data for control variables.39
Empirical specification We use a two-part model to estimate the effect of smoking by school personnel and school-level variables on youth smoking. In the first part, we model smoking participation for all students using a linear probability model. In the second part, we estimate cigarette consumption among smokers. The conditional demand estimation uses a generalised linear model (GLM) with a Gamma distribution and log-link.40 41 We cluster all standard errors at the school level.
RESULTS OLS results for smoking participation Table 2 summarises results for smoking prevalence. We only present the fully specified model. We ran several models (not shown) by including/excluding different controls, to make sure that our estimates were robust and did not change. The point estimate—0.063—of personnel smoking at school is highly significant. A 10-percentage point increase in smoking on school grounds by personnel is associated with an increase in smoking prevalence of 0.63 percentage points—a 5.7% increase in smoking prevalence. Smoking by school personnel is linked to a larger absolute effect on smoking prevalence among boys than among girls ( parameter estimates of 0.069 vs 0.05). However, the relative effects on smoking prevalence for a 10-percentage point increase in smoking among personnel are 4.6% and 7.1% for boys and girls, respectively. Nikaj S, Chaloupka F. Tob Control 2015;0:1–7. doi:10.1136/tobaccocontrol-2015-052531
Other controls have the expected signs. Males are more likely to smoke than females. Maternal smoking and paternal smoking are associated with increases in the probability of smoking by 11.1 and 4.6 percentage points, respectively, highlighting the large influence parents can exert on their children’s smoking. Greater exposure to cigarette advertising is associated with increased smoking prevalence. Exposure to antismoking media is linked with a decrease in prevalence, but the association is significant only at the 10% level. Being denied sale of cigarettes due to one’s age is associated with a reduction in smoking participation among girls. Instruction at school about the dangers of smoking is linked to a lower probability of smoking for all youths. On smoking restrictions at school, the parameter on having a student ban is insignificant or has a positive association (for females). The most likely interpretation is that schools with smoking bans often have a larger share of student smokers. In fact, schools often adopt such policies if smoking among the student body becomes a persistent or increasing problem, leading to a positive correlation between smoking bans and smoking among students. A second potential explanation may be that having a ban may not reduce participation rates unless schools enforce such bans comprehensively. The inclusion of an interaction between the enforcement variable and the student ban variable suggests the effect is zero. This finding should not be surprising, as our variables are likely capturing the effect of both policy selection—where schools with a large share of smokers are more likely to adopt smoking bans among students —and the effect of the ban. Since policy selection is positively associated with smoking participation, and bans and enforcement of such bans are negatively associated with smoking, the net effect is comprised of these two opposing effects, which could explain our null findings. A ban on smoking for personnel is associated with reductions in the probability that girls smoke (significant at the 10% level). Adding the interaction of enforcement and smoking bans for personnel does not affect the decision to smoke among youth. 3
Downloaded from http://tobaccocontrol.bmj.com/ on January 1, 2016 - Published by group.bmj.com
Research paper Table 2
Linear probability model of smoking participation
Staff smoking Student ban Staff ban
All†
Males†
Females†
All†
Males†
Females†
0.063*** (0.013) 0.019+ (0.011) −0.014 (0.009)
0.070*** (0.017) 0.016 (0.016) −0.011 (0.012)
0.050*** (0.012) 0.022* (0.010) −0.013+ (0.008)
−0.149*** (0.011) 0.006*** (0.000) 0.081*** (0.002) 0.111*** (0.008) 0.046*** (0.002) 0.120*** (0.006) 0.098*** (0.021) −0.012+ (0.006) −0.034 (0.021) −0.010*** (0.002) 229 995
−0.171*** (0.016) 0.007*** (0.001)
−0.120*** (0.013) 0.004*** (0.000)
0.069*** (0.017) 0.01 (0.022) 0.004 (0.027) 0.009 (0.024) −0.02 (0.033) −0.171*** (0.016) 0.007*** (0.001)
0.050*** (0.012) 0.02 (0.013) −0.015 (0.018) 0.002 (0.015) 0.002 (0.022) −0.120*** (0.013) 0.004*** (0.000)
0.116*** (0.012) 0.064*** (0.003) 0.138*** (0.009) 0.060* (0.031) −0.014 (0.010) −0.022 (0.033) −0.009*** (0.003) 115 284
0.105*** (0.008) 0.030*** (0.002) 0.103*** (0.007) 0.127*** (0.018) −0.008 (0.005) −0.037* (0.017) −0.009*** (0.002) 114 711
0.063*** (0.013) 0.021 (0.015) −0.01 (0.020) −0.001 (0.017) −0.005 (0.024) −0.149*** (0.011) 0.006*** (0.000) 0.081*** (0.002) 0.111*** (0.008) 0.046*** (0.002) 0.120*** (0.006) 0.098*** (0.021) −0.012+ (0.006) −0.034 (0.021) −0.010*** (0.002) 229 995
0.116*** (0.012) 0.064*** (0.003) 0.138*** (0.009) 0.060* (0.031) −0.014 (0.010) −0.022 (0.033) −0.009*** (0.003) 115 284
0.105*** (0.008) 0.030*** (0.002) 0.103*** (0.007) 0.127*** (0.018) −0.008 (0.005) −0.037* (0.017) −0.009*** (0.002) 114 711
Student ban enforcement Staff ban enforcement Age Age squared Male Mother smokes Father smokes Both parents smoke Exposure to cigarette ads Exposure to antismoking ads Reduced access Class on smoking N
All specifications include country fixed effects, and linear and quadratic time trends. +p
