Abstract

FERNANDEZ, Paula; VALLEJO, Guillermo  and  LIVACIC-ROJAS, Pablo. Robustness of five univariate statistics in analyzing Split-Plot desings under adverse conditions. rev.latinoam.psicol. [online]. 2010, vol.42, n.2, pp.289-309. ISSN 0120-0534.

In this research we examine the behaviour of five univariate statistics for analyzing the data of a Split-Plot design. Four of them assume that the dispersion matrix underlying is not spherical. However, they do so with a clear distinction between two alternatives, insofar as two of them presuppose that the correlation between the data does not have a certain structure and other two assume that there exists first-order serial autocorrelation. All of them were compared with regard to their robustness to test the sources of variation within-subject (treatment and interaction) under non-normality in the absence of sphericity, both when there was first-order serial autocorrelation and when the underlying correlation was arbitrary. The results show that when the distribution is non-normal symmetric all the procedures show a Type I error rate similar to the obtained one under normal distribution. As the degree of skewness and kurtosis increases, all the procedures experience an alteration in their estimation of the Type I error rate and that it depends on the structure of covariance matrix underlying in the data. In the set of conditions submitted to study the most robust procedures were HCH, JN and LEC.

Keywords : Robustness; first-order serial autocorrelation; absence of sphericity; absence of normality.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )