Losing significance from composite variable to factor score

[ksmielitz]

I have two variables that are supposed to become one DV. This is not a problem, as they are both scales and I had planned on doing a summation and running an OLS. Then, due to some discussion with my major professor, we decided to confirm with a factor analysis, to make sure they loaded together was appropriate--fine, no big deal. Well, now we have the issue.

If I run them as a composite variable 

finmanagecomposite=sum(of likelysave likelybudget);

proc reg;
model finmanagecomposite= SES priorfined black finknow subfinknow aspectincar yesacct committed adult late30s olderfolk money71 pbc attitude violent truth compareself outsidesubnorm/vif tol;
run;

I get the results I would expect with my predictors of interest (per theory) achieving significance in predicting my DV.

However, when I run the factor score: 

proc factor data=dissert score outstat=factout nfactors=3
		method=prin rotate=varimax score;
	var open likelyafs likelybudget likelysave totaltime numincar;
run;

proc score data=dissert score=factout out=fscore
	(rename= (factor1=bankincl factor2=finmanage factor3=AspectIncar));
	var open likelyafs likelybudget likelysave totaltime numincar;
run;

***for clarity I'm only worried about factor2 above)***

proc reg;
model finmanage= SES priorfined black finknow subfinknow aspectincar yesacct committed adult late30s olderfolk money71 pbc attitude violent truth compareself outsidesubnorm/vif tol;
run;

I lose my significance. 

Why? Shouldn't the composite variable (1st section of code) and the factor score (immediately above) result in the same general description of my DV?

Thanks in advance, 

Kate

[PaigeMiller]

The response variable in your first section of code is not the same as the response variable in the 2nd section of code. So why should they produce the same results?

Shouldn't the composite variable (1st section of code) and the factor score (immediately above) result in the same general description of my DV?

This is rather a bold statement, how would you really know? And it appears to be contradicted by the data. If it was true, if you could somehow KNOW that variable finmanagecomposite is essentially the same as variable finmanage, then why do the statistical analysis at all?

As a side issue, I find myself unclear as to how factor analysis helps here, to me it could harm the situation just as much as it could help — it could be adding noise to the variable finmanage that is not present in finmanagecomposite, as factor analysis is not designed to pick composite scores that are well predicted by your x-variables.