Dealing with hierarchical data in periodontal research

Abstract

Site-specific clinical periodontal data are usually plentiful, typically hierarchical, and generally valuable information. Summarizing these data on a subject level for easy application of standard statistical tests leads to loss of most of the information. In addition, well-known fallacies may make interpretation difficult if not impossible. In this study, an attempt is made to apply, in a non-technical way and as a tutorial, a rather complex multilevel model of gingival thickness, which provides unbiased estimates of fixed effects and a variance/covariance matrix with considerable information as regards data structure. When applying multilevel modeling, random effects should generally be reported in a proper way, since they might reveal new insights into subject and tooth variation, correlations between covariates, and even problems with the chosen model.