Variance components analysis of data from periodontal research.

Although periodontal disease status and progression may vary at different sites in the same subject, statistical models of data from periodontal research must not assume independence of different observations made on the same subject. Methods to analyze appropriate statistical models for data from periodontal research exist, but have hitherto required the data set to be generally balanced. Algorithms for the estimation of `variance components' or `multilevel' models for the analysis of unbalanced data which possess a nested structure have, however, recently been developed. The use of a simple variance components model is illustrated using data from a periodontal clinical trial. The conclusions reached differed from those reached by the original analysis, which did not take account of within-subject differences. Such models may be generalized, for instance to allow for random effects or Longitudinal data. These models will be useful in the study of the etiology and pathogenesis of periodontal diseases, [ABSTRACT FROM AUTHOR]