MANOVA for different dimensions of the rubric

[Cohen]

We have two groups of students divided into control and experimental group, with a specific teaching method being the intervention.
There are about 35-40 students in each group.
We are grading one of their assignments using a rubric which has 4 dimensions (or 4 parts). So each student will get a score for each of those dimension AND a total score on their assignment. We want to test the difference between scores for experimental vs. control group to see if the intervention had any effect.

I am planning to use a MANOVA test using the grades for the four dimensions/parts of the rubric as dependent variables.

1. Does this approach make sense?
2. Can I also include the total score as one of my dependent variables (so have 5 dependent variables)? or that doesn't make sense?

Thanks!

 

[fed2]

i think yes on 1) and no on 2)

I think 2) will cause the MANOVA to explode, just instinct though.

 

[Cohen]

I guess my question is that does including total score as one of the dependent variables conceptually make sense?

 

[fed2]

it makes 'conceptual sense'. However, This type of 'multicollinearity' among the dependant variables is bad, if not outright forbidden in almost-surely all stats models.

for the sum probably easiest approach is to just run a separate one-way anova.

It may also be possible to implement as a contrast in the MANOVA. See for example
https://webpages.uidaho.edu/CALS-StatProg/sas/workshops/multivariate/manova.pdf
especially the treament of 'Total Setback'.

hope that helps in your manova quest.