Add to Calendar
When:
March 12, 2012 @ 1:00 PM – 2:00 PM
2012-03-12T13:00:00-04:00
2012-03-12T14:00:00-04:00
Where:
Norm Endler Seminar Room (BSB 164)
Cost:
Free
Speaker: Carrie Smith, York University
Department of Psychology
Title: Multivariate analysis of variance (MANOVA) under heteroscedasticity and non-normality.
Abstract: In many practical situations it is necessary to compare 3 or more groups of subjects on the means of several dependent variables. The problem of comparing means for multiple outcome variables (mean vectors) is known as the multivariate analysis of variance (MANOVA). The most commonly employed MANOVA tests (e.g. Hotelling's T^2 (1931), Roy’s largest root (1954), Hotelling(1951)-Lawley(1938) trace, Pillai(1939)-Bartlett(1955) trace and Wilks’ lambda (1932)) assume equivalence of population covariance matrices and multivariate normality, assumptions that are routinely violated in applied behavioural research.
In this talk I will describe a simulation study conducted to assess performance of three procedures that do not assume equivalence of covariance matrices: Johansen's MANOVA (Johansen, 1980), Johansen's MANOVA with trimmed means (Wilcox, 1995) and a parametric bootstrap MANOVA (Krishnamoorthy & Lu, 2010). The aim of the study is to identify procedures that maintain Type I error rates below the desired nominal level (alpha) while yielding maximal statistical power under varying degrees of heterogeneity of variance and multivariate non-normality (both skewness and kurtosis). Preliminary results and recommendations will be discussed.
