Really a Likert-scale (which consists of several Likert-type items), or just one Likert-type item?
The latter is ordinal scaled.
Generally speaking, as far as I know MANOVA is particularly suitable if the dependent variables
jointly represent a hypothetical construct, and if the variables are neither correlated very high
or very low. In the present case, I do not know whether "support for affirmative action" and
"hiring a particular (?) woman" jointly represent one construct. Moreover, the formulation
"the negative effect of X on A decreases B" seems to indicate a mediation model, not a