Multilevel Model In Epidemiological Studies Using Asthma As An Example

Multilevel models are increasingly being used to analyze clustered data in public health, epidemiological and educational research. This article introduces the basic knowl-edge about multilevel model, including basic form, methods of parameter estimation and sample size problem. Two-level Logistic Multilevel model is applied to national survey data in2005to explore the relationship between asthma and the individual indicators and urban indicators. Studies have shown that suffering from asthma is primarily caused by individual physical indicators, such as sex, suffering from colds frequently, ever diagnosed allergies and so on, meanwhile the factors from the city cannot be ignored, especially average humidity. The results also show that the multilevel Logistic model’s goodness of fit is better than single Logistic model which is reflected at the smaller log-likelihood values and residual standard errors. In addition, this article shows a Monte Carlo study to compare the performance of different situations (the combination situations of the number of cluster, the number of subject per cluster as well as Intra-Class Correlation Coefficient(ICC)) based on the low probability of response variable. And it tells us that the number of cluster is more important than the number of subject per cluster; most of estimations tended to be poor when there were only five subjects per cluster, regardless of the number of clusters,especially for the higher ICC situation.