Practical issues

Question 1

How many DVs?

Answer 1

DVs need to have a strong (empirical/ theoretical) rationale

Question 2

Problem with large number of DVs

Answer 2

:( power of a multivariate test generally declines with increased number of DVs

Small difference in most of them can obscure a real difference in 1 or more DV

Multivariate tests detect mainly error variance so declares no overall difference

Question 3

How is sample size estimated?

Answer 3

1) some say 20 minimum per group
2) some say number IV levels x number of DVs per group
3) use gpower to calculate required sample size for power .8 and a=.05, with 3-6 DVs (based on epected effect size f-squared e.g. .05)

Question 4

How are multivariate hypotheses distinct from univariate ones?

Answer 4

e.g. IV= treatment
DV: academic performance (DV1: reading, DV2: maths)

UV hypothesis: does treatment have an effect on reading performance? does treatment have an effect on maths performance?

MV hypothesis: does treatment have an effect on intellectual ability (e.g IQ or another suitable label)

Question 5

Manova null hypotheses

Answer 5

Main effect of IVs:
IV has no systematic effect on the optimal linear combination of DVs.

Interactions among IVs:
Change in the LC over levels of IV does not depend on the level of another IV.

Question 6

manova research quetions

Answer 6

1) Importance of DVS:
Which of the DVs are changed and which are unaffected by the IVs? (this shades into univariate territory)

2) Specific comparisons:
Which levels (groups) of IV main effect are different from which others?

3) Effect sizes:
How large are the effects? What proportion of the variance of the LC is attributable to the effects? (partial eta squared)