Rerandomization tests for analyzing correlated data from dental studies

✍ Scribed by Mark A. Espeland; William C. Murphy; Christopher Cox; Ronald J. Billings; Otis J. Bouwsma

• Publisher: Elsevier Science
• Year: 1989
• Tongue: English
• Weight: 662 KB
• Volume: 22
• Category: Article
• ISSN: 0010-4809
• DOI: 10.1016/0010-4809(89)90009-8

✦ Synopsis

Dental research studies often produce relatively small data sets in which observations are serially or spatially correlated. Rerandomization tests are presented as alternatives to analysis of variance and multivariate analysis for assessing group differences using such data. Rerandomization tests are particularly useful when the investigator is unwilling to make strong assumptions about the nature of the serial correlation or the distribution of the data. Two examples are discussed that demonstrate these techniques. 8 1989 Academic press, IIIC.

Dental research studies often generate repeated measures on subjects or experimental units. For example, each group of subjects may be examined at several longitudinal time points or data may be collected simultaneously at a number of sites in the mouth. While it can be useful to condense such data to single, overall statistics, such as an average across time or a whole mouth score, often it is important to examine patterns among time points or sites. To do such an analysis correctly, one must take into consideration the correlational structure linking the repeated measures. Although it is not uncommon to find in the literature examples where repeated measures have been analyzed independently, such as performing series off tests at each time point or site, the results of ignoring correlations can lead to spurious conclusions (Cohen and Cecil (I)).

Typical methods for analyzing repeated measures data are multivariate analyses of variance (e.g., Fertig et al. ( )), growth curve methods (e.g., Rao (3), or