Two DV's in a 2x2 factorial experiment I'm trying to investigate the effect on (a) tear resistance and (b) gloss of a piece of film due to additive and extrusion method. I can see that the best results for tear resistance is low additive high extrusion, and the best results for gloss is low additive, low extrusion. Although is there a way to say the best overall method - i.e. for best gloss AND tear resistance? Or a way to see if there is an effect between the two DV's - i.e. will the glossier a piece of film is mean that there is less tear resistance?
Would a scatter plot with no correlation confirm that there is no effect between gloss and tear resistance?
I want to be able to have an overall recommendation stating that the best overall results occurr when additive is at 'x' level and extrusion at 'y' level?
I also cannot verify normality (as there are only 3 replications) or homogeneity of variance as SPSS states that there are fewer than two non singular cells.
Any help would be much appreciated...
 A: The optimum combination is inherently subjective because the best compromise between tear resistance and gloss is inherently subjective.  I've seen several approaches.  One is to superimpose contour plots.  It is described in Box, Hunter & Hunter, "Statistics for Experimenters".  Another is the desirability function approach popularized by George Derringer.  Both are built into JMP software. Yet another might be to create a grid of predicted tear resistances and glosses, plot predicted tear resistance vs predicted gloss and pick the combination you like best. 
A: There multivariate methods that would allow you to look for a joint impact of multiple independent variables on multiple dependent variables. The particular one that you'll use depends partly on the nature of the dependent variables.
If you have some concept of the relative importance of tear resistance gloss, you could combine them into one variable. For example, if tear resistance is twice as important as gloss, then maybe $y=2\times tear\_resistance+gloss$.
But I think you're going to have trouble testing anything with such a small sample ($n=12$).
