NEW EXAM RM MANOVA (PROFILE ANALYSIS) Flashcards
What is a design with multiple DVs measured at multiple time points, but the multiple DVs do not have to be measured on the same scale?
Doubly Multivariate
Outline the 4 similarities between repeated measures analysis of variance (ANOVA) and repeated measures MANOVA
1) address the same questions
2) test the same effects in the design (main effects and interactions)
3) applied to the same (repeated-measures) designs.
4) They are identical when the repeated-measures factors have only 2 levels.
Highlight the 2 differences between repeated measures analysis of variance (ANOVA) and repeated measures MANOVA
Differences occur for any repeated-measures factor with >2 levels and they relate to:
1) assumptions: in particular RM ANOVA requires compound symmetry of the variance-covariance matrix often tested via sphericity (homogeneity of covariance) whereas RM MANOVA does not
2) test statistics: there are 4 for RM MANOVA (Wilks lamda etc) whereas only 1 (F-ratio) for RM ANOVA
Profile analysis is a special application of multivariate analysis of variance (MANOVA) to a situation where
there are several dependent variables (DVs), all measured on the same scale.
The set of DVs can either come from 2 routes
1)
one DV measured several different times -
(so between groups factor measured on DV1 time 1 - DV2 time 2 - DV3 time 3)
2)
or several different DVs all measured at one time (baseline - treat - post - baseline - treat - post = all on one composite DV)
The choice between profile analysis and univariate repeated-measures ANOVA depends on 3 main things….
1)
sample size
2) power
3)
whether statistical assumptions of repeated-measures ANOVA are met.
what 3 hypotheses are tested?
flatness
levels
parallelism
Main limitation to profile analysis
1)
Choice of DVs is more limited in profile analysis than in usual applications of multivariate statistics, because DVs must be commensurate except in the doubly multivariate application. That is, they must all have been subjected to the same scaling techniques.
Way round is Z score DVs - but…
There is some danger in generalizing results with this approach, however, because sample standard deviations are used to form z-scores.
In the choice between univariate repeated-measures ANOVA and profile analysis, xxxxxx xxxxxx is often the deciding factor.
sample size per group is often the deciding factor.
The sample size in each group is an important issue in profile analysis, as in MANOVA, because there should be more research units in the smallest group than there are DVs. This is recommended both because of considerations of power and for evaluation of the assumption of homogeneity of variance–covariance matrices
what is not an issue with profile analysis ?
Unequal sample sizes typically provide no special difficulty in profile analysis because each hypothesis is tested as if in a one-way design and, as discussed in Section 6.5.4.2, unequal n creates difficulties in interpretation only in designs with more than one between-subjects IV
what is the rule of thumb from terbachnick about sample sizes?
there should be more research units in the smallest group than there are DVs.
If it is small, use RM ANOVA
If sample sizes are equal, what is not necessary?
If sample sizes are equal, evaluation of homogeneity of variance–covariance matrices is not necessary.
what are the two main advantages of RM ANOVA?
1)
Reduced error (within-group) variance (P’s function as own control group – e.g. perhaps placebo condition were just better on that task)
2)
Statistical power increased – fewer P’s needed.
what are the three main limitations or concerns of RM ANOVA?
1)
Order effects
2)
Carry over effects - A carryover effect is an effect that “carries over” from one experimental condition to another. Whenever subjects perform in more than one condition (as they do in within-subject designs) there is a possibility of carryover effects.
3)
Sensitisation - sensitised to treatment for instance
one main limitation of profile analysis given by Devin?
Scaling – Limited as DV has to be on same scale (can use Z scores, but harder to interpret)