Week 10 Flashcards
Do DV’s in MANOVA have to be conceptually related?
Yes.
Typically they are multiple ways to measure a concept.
What is a benefit to using MANOVA?
Lower type 1 error rate than calculating several univariate ANOVA’s for each DV.
For analysis via MANOVA, the DV’s:
- Should be conceptually related
2. should be moderately correlated
If you are planning to do a MANOVA but the DV’s are not correlated, what should you do instead?
Separate ANOVA’s with Bonferroni adjusted p values.
If you are planning to do a MANOVA and the DV’s are too highly correlated:
There is a redundancy of information which loses power.
What are two reasons that a MANOVA might lose power due to highly correlated DV’s?
- multicollinearity
2. Singularity
What is multicollinearity?
- Very high correlations between variables
What is singularity?
One variable is a combination of a number of others in the analysis.
At minimum, in a MANOVA analysis, the number of DV’s should be:
Less than the number of cases.
What are the assumptions for MANOVA?
- independance of observations
- multivariate normal distribution (large sample sizes help with this)
- homogeneity of variance for all groups
MANOVA has to be what type of subjects?
Between
Homogeneity of variance in a MANOVA is usually okay if there are an equal number of participants in each cell. But, to be sure, what are two ways we can check this?
- Levene’s test
2. Box’s test
Describe Box’s test
very sensitive to violations of multivariate normality so pay little attention to it unless unequal sample sizes (or p
If you have a significant Box’s test, what’s one thing you can do after this?
Do a Levene’s test to see where the problematic DV might be
What is the default multivariate test of significance we should use in a MANOVA?
Wilks Lambda - generally recommended unless some possible assumption violations
Which multivariate test of significance should we do for a MANOVA of the design is flawed (small n, unequal numbers etc) or is we have a significant Box’s test?
Pillai’s Trace.
Wilks Lambda is equal to what in regard to variance?
Unexplained variance. So, variance accounted for by the best linear combination of DV’s (aka the MANOVA) = 1 - Wilks Lambda.
This is how we can work out the effect size for MANOVA.
Pillai’s V is what in regard to effect size?
The proportion of variance accounted for by the best linear combination of DVs.
What do we do if we have a significant MANOVA?
- separate univariate tests of all DV’s. One ANOVA DV at a time.
- step down analysis using ANCOVA.
What happens if the MANOVA is not significant?
Differences here may be type 1 errors. Safest thing to do is report there was no significant differences and don’t do any further analysis.
MANOVA _____ be more sensitive at detecting differences between IVs.
MAY
What is preferable to have in data with MANOVA?
Preferable to have large samples, equal sample sizes and no outliers.
