Statistical analysis of repeated measures in psychiatric research

We examined the statistical methods applied to research studies that used parallel-groups repeated measures designs. Sixty-three of the 343 articles that were published in four major psychiatry research journals during a 6-month period contained data of this type. Seven (11%) of the 63 articles applied a multivariate analysis of variance. Four (6%) used a univariate analysis of variance with an adjustment for correlated measures. Ten (16%) used a univariate analysis of variance without an adjustment for correlated measures. Twenty (32%) either reduced their repeated measurements to a single value, such as a maximum-change score in each group, or carried out multiple tests at several time points. Twenty-two (35%) of the reports did not provide enough information to allow readers to determine the type of analysis that was applied. We discuss the appropriate selection of analytical techniques for repeated measures designs.