Setting and Sampling Design
The study was conducted in the Province of Jujuy in northwest Argentina where about 44% of the population lives below the poverty level and the unemployment rates in recent years reached 25% (INDEC, 2001
Secondary schools were randomly sampled from within the three geographic areas of the province: the mountain region; the provincial capital; and the agricultural lowlands. Secondary schools include 8th through 12th grades and reflect the standard private and public school organization in Argentina. Based upon government data in 2004, we estimated the number of 8th grade students within each region to equal 1,509, 7,296, and 5,379, respectively. The goal was to select a representative sample of schools containing approximately 1,000 8th grade students from within each geographic area (i.e., disproportionate stratification). The original sample of schools included 9 of the 17 schools within the mountain region, 6 of the 48 schools in the capital, and 9 of the 44 schools in the lowlands. The original set of schools was augmented by adding two schools in the capital region to over-sample students from poor urban areas. Furthermore, in the lowlands, two schools declined participation and were replaced by three additional schools from the same area. The final sample included 27 schools, 3 of which were private.
We worked with the provincial government office that monitors substance and alcohol use. The selected schools were informed of the planned study and the school principals’ cooperation was solicited. We provided modest incentives to the schools in the form of supplies and training. There was no monetary compensation to youth or schools for collaboration. Spanish-language surveys were administered sequentially to schools between June and August of 2004, and within each sampled school, we attempted to recruit every enrolled 8th grade student into the study. All youths spoke Spanish and schools were conducted only in Spanish. School coordinators were trained on administration of the survey and trouble-shooting questions during the session. Surveys were self-administered in class with research staff and school coordinators present as proctors. In each school, one attempt was made to survey absentee students on a subsequent date.
The research protocol was approved by the UCSF Committee on Human Rights and by an NIH-certified human subjects research board in Buenos Aires based at Centro de Educación Médica e Investigaciones Clínicas (CEMIC). Passive consent was requested from parents or caretakers and students signed an active consent form to allow follow-up contact for subsequent surveys.
Development of Questionnaire Measures
The questionnaire included translations of items used in surveys of adolescents in the US (Global Youth Tobacco Survey Collaborative Group, 2002
) and items developed from our previous qualitative research. Items in English were translated and back translated and reviewed by three Argentinian investigators and two other Spanish-speaking research staff. Pilot testing of the instrument was conducted with students in rural and urban areas evaluating situational factors, content, characteristics of the respondents and time of administration (average of one hour). Instructions were developed to address the importance of providing accurate answers and the confidential nature of the survey.
Outcomes: Smoking Behavior
Smoking behavior questions were developed to be comparable to those used in the Centers for Disease Control and Prevention GYTS survey (Global Youth Tobacco Survey Collaborative Group, 2002
) and included age at smoking initiation, number of cigarettes smoked in the lifetime, in the past 30 days and in the past week, and how many days in the past month and past week respondents smoked. Respondents were considered ever smokers
if they tried at least a cigarette puff in their lifetime and never smokers had not tried even one puff. Current smokers were defined as having smoked at least one whole cigarette in their lifetime and at least one puff in the previous 30 days. Established smokers were defined as current smokers who had smoked at least 100 cigarettes in their lifetime. In order to address ceremonial use of tobacco among respondents who reported ever smoking, we asked in what place or situation they had first tried smoking and where they felt like smoking the most; responses that mentioned the Pachamama and Todos Santos traditional celebrations were included.
Exposure Variables: Demographics, Family and School Characteristics
Exposure or predictor variables included demographic, family and school characteristics. Respondents reported their sex, date of birth, age, religion, and if they were currently working. Ethnic self-identification was ascertained by providing a list of five ethnic categories to choose from: Indigenous, Mixed Indigenous and European (referred to as Mixed), European, Arab, and Other. Prior to including the ethnicity questions in the survey we completed qualitative work to evaluate whether youth were able to identify and distinguish among these ethnic categories. Family characteristics included parent's formal education and employment status, number of parents in the household (two vs. one or none), and family members speaking an indigenous language. We ascertained school location (rural, small town, urban), shift (morning, afternoon, evening) and type (public or private).
Exposure Variables: Psychosocial Risk Factors
Students reported on the presence of smokers in their household whether they ever repeated a grade in school, drinking at least one glass of alcohol in the previous week, and the number of their friends who smoked. Depressive symptoms were ascertained by asking a single item on whether the respondent in the past year felt sad and could not carry on his/her normal activities or obligations for at least two weeks (Benjet et al., 2007
). The thrill-seeking orientation measure was adapted from one used with youth living in Florida and was measured with three items using a five-point disagreement-agreement response set: “I do not mind getting in trouble as long as I have fun”; “I like to do dangerous things”; “I like to do things that people say should not be done.” The measure had a range of 1 to 5 and we defined scores of 3 to 5 as high on thrill seeking, and less than 3 as low on thrill seeking and thus adapted the measure to use as a binary version of the scale (Vega, Warheit, Apospori, and Gil, 1993
Expired Carbon Monoxide Validation
To evaluate potential under-reporting we measured expired air carbon monoxide (CO) using the Micro Medical Micro CO Meter. Respondents with 10 or more parts per million were considered to be current smokers. Expired CO was measured in a probabilistic sample of 452 respondents from one school in each geographic region of the province. The sample was selected proportional to school size.
The sampling design was incorporated into all models by specifying geographic areas as strata and schools as clusters, as well as including weights to adjust for disproportionate stratification. In addition, a finite population correction was applied to adjust for the relatively large proportion of available schools sampled within each geographic area. Standard errors and confidence intervals were estimated via the Taylor expansion approximation (SAS, 2006
). Because the data contained missing values, each substantive model was fit to 20 multiply imputed data sets created with SAS PROC MI (SAS, 2006
). To accommodate the sampling design, separate imputation models were fit to the data from each geographic stratum, and each stratum-specific model included a set of indicators representing the sampled schools, or clusters (Reiter, Raghunathan, and Kinney, 2006
). Imputed values for binary and categorical variables were rounded and truncated to the nearest category (Allison, 2002
; Schafer, 1997
). All parameter estimates and significance tests were calculated by combining results across the imputed 20 data sets (Meng and Rubin, 1992
; Rubin, 1987
Primary research questions included estimating total smoking prevalence, as well as within and across demographic groups defined by sex, age, and ethnicity. First, we conducted descriptive analyses to profile the sample examining the distribution of demographic, family and school characteristics, and psychosocial risk factors by ethnicity. Chi-square tests and p values were calculated. Bivariate contingency tables examined the pair-wise relationships among smoking outcomes, sex, and ethnicity.
Multivariate logistic models regressed the smoking outcomes (ever smoker, current smoker, and established smoker) onto demographic (sex, age, ethnic identity, currently working), family (parental education and employment status, number of parents in the household), school characteristics (location, type, and shift), and psychosocial risk factors (presence of smokers in the household, number of friends who smoke, repeating a grade, alcohol drinking, depressive symptoms, thrill seeking orientation). We estimated adjusted odds ratios and 95% confidence intervals.
For the “ever smoked” outcome, we additionally evaluated possible interactions between ethnicity and each risk factor. The modeling approach included all interactions and then removed the nonsignificant interactions (p > .05) via a backward elimination procedure, but all main effects were retained in the model.