Dear All,
It is entirely possible that I am thinking too much about this (I'm hoping)...but here's the DL.
I have conducted multiple factor analyses. In FA 1 I ran open, likelyafs, likelybudget, likelysave. Open and LIkelyAFS very nicely loaded together (expected) as did likelybudget and likelysave (also expected). So, moving on through my data and coding I also needed to see how totaltime, numincar, and offense loaded so I could determine if I have to leave them as separate predictors or if I could create a factor score. Totaltime and numincar very nicely loaded together and offense is hanging out there in the wind (that's fine).
So far, life is good.
I created the factor scores for open/likelyafs (bankincl) and likelybudget/likelysave (finmanage). Great...they work well! These are my DVs and I'm cruising along.
I'm working with a smaller sample (ah the joys of primary data) so I'm trying to keep my predictors to a minimum and said "AH! Since totaltime and numincar loaded so well together (.80, .82) on the same factor, I'll create a factor score for those as well." So, I did the exact same coding procedure I did for my DVs. But, when I run both proc factor/proc score combos (one for my DVs and one for my Aspects of Incarceration), my regressions won't run. HOWEVER, when I load them all into the mix together, the regressions run. This is what I have:
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;
FYI (a little more back story): For the sake of trying to figure things out, I ran a proc factor with everything together:
proc factor rotate=varimax ev scree min=1;
var totaltime numincar open likelyafs likelybudget likelysave;
run;
And boy howdy did the factors show up whacked. I was *hoping* that they would separate, but alas, we can't have everything can we? So, I deleted that code and said, "Well, I have 3 clearly defined factors when I run the proc factor by the general subjects (DVs and aspects of incarceration), so I'll use those and load them into the proc factor/proc score to get everything labeled since trying to score and label everything separately blocks my DV factor scores from running (i.e. SAS says they don't exist)."
i.e. when I do the following my DVs "do not exist":
proc factor data=dissert score outstat=factout nfactors=1
method=prin score;
var totaltime numincar;
run;
proc score data=dissert score=factout out=fscore
(rename= (factor1=AspectIncar));
var totaltime numincar;
run;
proc factor data=dissert score outstat=factout nfactors=2
method=prin rotate=varimax score;
var open likelyafs likelybudget likelysave;
run;
proc score data=dissert score=factout out=fscore
(rename= (factor1=bankincl factor2=finmanage ));
var open likelyafs likelybudget likelysave ;
run;
proc reg;
model bankincl=age finknow aspectincar;
run;
proc reg;
model finmanage=age finknow aspectincar;
run;
When I run my regressions with the proc factor/proc score in the first example of this novel:
proc reg;
model bankincl=age finknow aspectincar;
run;
proc reg;
model finmanage=age finknow aspectincar;
run;
They run...but since this is my dissertation I have a particularly vested interest in making sure that there is nothing wrong with what I've got so far.
With all of that information, it boils down to: Is there anything wrong with loading all of my factors into one scoring mechanism? I think there isn't because my DVs are separate factors but I'm *minorly* freaking out. 🙂
Thanks for sticking with me through the novel...maybe my dissertation will be as long. 😉
Kate
