[An applied multilevel model used in the study on behavior of smoking]

To introduce the nature and its application of a multilevel model.

Data was analyzed from a baseline survey of smoking behavior among middle school students sponsored by a WHO smoking controll project. Multilevel analysis was used on available data and to compare the results from logistic regression.

The outcomes of null multilevel model approved that there was hierarchical structure on data. The influencing factors of middle school students smoking appeared to be gender, age, attitude, environment and public education at schools. When the variable of public education by classes was not included, the standard errors by multilevel analysis became smaller than the corresponding standard errors through logistic regression method.

Multilevel model seemed a good method for analyzing data with hierarchical or cluster structure, it could be applied in stratified or cluster sampling of epidemiological or community-based investigation.