Introduction to Multilevel Analysis

Abstract

In this era of large-scale, multi-institutional studies, the importance of analyzing hierarchical (clustered) data is increasing. However, conventional regression analysis may be inadequate for this purpose because it assumes that records for individual patients are independent of records for other patients. Multilevel analysis is a statistical method that allows one to analyze data with a hierarchical structure. Mixed-effect models expand the conventional regression models by incorporating random coefficients for each unit of cluster. Multilevel analysis can be applied to studies on repeated measures among individuals, multi-institutional studies on effects of individual-level variables, and studies on effects of cluster-level variables or cluster-level variances. This report summarizes the basics of multilevel analysis and how it can be used in clinical epidemiology research.