Abstract
Objectives: This study was conducted to estimate the unbiased smoking prevalence and its standard errors among adolescents in a large city in Korea, by design-based analysis. Methods: All the students in Daegu city were stratified by grade, gender and region, and then schools as primary sampling units (PSU) were selected by probability proportional to size (PPS) sampling. One or two classes were sampled randomly from each grade, from 5th grade in elementary schools to the 3rd grade in high schools. The students anonymously completed a standardized self-administered questionnaire from October to December 2004. The total number of respondents was 8,480 in the final analysis, excluding the third graders in the general high schools because of incomplete sampling. The sampling weight was calculated for each student after post-stratification adjustment, with adjustment being made for the missing cases. The data were analyzed with Stata 8.0 with consideration of PSU, weighting and the strata variables. Results: The smoking prevalence (%) and standard errors for male students from the fifth grade in elementary schools to the second grade in high schools were $0.93{\pm}0.47,\;1.83{\pm}0.74,\;3.16{\pm}1.00,\;5.12{\pm}1.02,\;10.86{\pm}1.13,\;15.63{\pm}2.44\;and\;17.96{\pm}2.67$, and those for the female students were $0.28{\pm}0.28,\;1.17{\pm}0.73,\;3.13{\pm}0.60,\;1.45{\pm}0.58,\;3.94{\pm}0.92,\;8.75{\pm}1.86\;and\;10.04{\pm}1.70$, sequentially. Conclusions: The smoking prevalence from this study was much higher than those from the other conventional studies conducted in Korea. The point estimates and standard errors from the design-based analysis were different from those of the model-based analysis. These findings suggest the importance of design-based analysis to estimate unbiased prevalence and standard errors in complex survey data and this method is recommended to apply to future surveys for determining the smoking prevalence for specific population.
