What type of test? Was going to use regressions but am now thoroughly confused.

[Playingwithnumbers]

So in the psych literature, passion can be divided into harmonious (HP) and obsessive passion (OP). Obsessive passion predicts high identification and identity conflict while harmonious passion predicts low identification and harmonious identity. So I have survey results with various measures of identity and measures of obsessive and harmonious passion. I am trying to compare OP and HP relationships with various identity measures. I was doing multiple regressions, with OP and HP as IVs and Identity as DV. But the problem is OP and HP are highly correlated (.6), so I think regressions are no go right? Should I use correlations instead?

Also, how do I study if the effect sizes of OP and HP on Identity are significantly different?

One choice I am trying to deal with whether I should just study the correlations between the OP construct, the HP construct and Identity, or split my sample into two groups, one high in HP, one high in OP, and do a...t-test? :\

Thanks a lot!

[CowboyBear]

Try obtaining Collinearity Diagnostics from within the regression dialog box; a rule of thumb is that the Variance Inflation Factor (VIF) should be under 4.0 for each IV. A correlation of 0.6 between two IV's is not necessarily the end of the world.

It strikes me, though, that it's worth considering what the correlation of 0.6 tells you about whether it's really possible to define two completely separate dimensions of passion (insofar as it's possible to measure something like passion at all).

[Playingwithnumbers]

Oh, thanks! My VIF is around 1.5. Hm.

Well, the psychology literature defines passion as such because while it correlates with variables that define passion (valuation, interest, time, etc), they differ significantly on what happens when the person does, or finishes, the activity (positive/negative affect; rigid/flexible participation, etc).

Thanks!

[trinker]

You could enter a highly correlated predictors set into regression (hierarchial fixed order) as a block and treat them as a single variable (Kind of)

[CowboyBear]

Oh, thanks! My VIF is around 1.5. Hm.

That's pretty low; it means the variance for that particular IV's coefficient is inflated by only a factor of about 1.5 due to collinearity. If multiple regression is a better conceptual fit for your study than correlation, sticking with regression may be fine. Make sure you look at all the other regression assumptions though.