ERIC Number: ED229389
Record Type: Non-Journal
Publication Date: 1981
Reference Count: N/A
A Simulation Comparison of Univariate and Multivariate Analyses of a Multi-Factor Repeated Measures Design.
Berger, Dale E.; Selhorst, Susan C.
Although it is widely known that special assumptions are needed for univariate analysis of repeated measures data, researchers seldom examine their data for violation of these assumptions. This paper reviews ways in which repeated measures analyses are usually handled and describes limitations of these methods. A design with two within subject factors (3x3) was tested with a computer simulation of 1,000 such experiments (each with 30 subjects) to examine the bias of alternate test procedures with data similar to that which might reasonably be observed. Two data structures were used, with small and large violations of the univariate assumptions. Four methods of analysis were compared: unadjusted univariate, Geiser-Greenhouse conservative test, epsilon correction, and multivariate analysis. The multivariate test was the only procedure for which the empirical alpha error rate did not differ reliably from the nominal alpha for any effect tested here. It is recommended that multivariate procedures should be used for analysis of repeated measures designs when sample size permits. (Author/PN)
Publication Type: Reports - Research
Education Level: N/A
Authoring Institution: N/A