Working through a messy EFA (factor analysis) in SPSS

first off I see we have some nominal scales here they can't truly be nominal can they as strongly disagree to strongly agree so let me just change these over to scale Oh quick and this one isn't matter we won't be using that the first thing I want to do is actually not affect analysis first thing I want to do is look at the reliabilities so I'd go to analyze scale reliability and just see if there's anything that doesn't fit very well within each of these factors so I'm gonna stick those 4ac in there I'm gonna go to statistics and check the box for scale if item deleted and they okay okay and see what we end up with 828 it's pretty good is there a way to get better mmm not really so we'll keep all those items for now we're gonna do the same thing again or the other ones number we're looking there before there is anything above 0.7 so that looks really good for AC let's go to NC and we have 753 not bad and no way to really improve that so the next one Oh C's CC okay 741 is pretty good and again no way to improve that so the next thing I want to do is go look at the wording here we go here's the wording of these items I assume these run strongly agree to strongly disagree yeah the opposite I mean we disagree or strongly agree good and there's no like not applicable which is good so this organization I mean I look at these ones here and make sure they're reflective they seem to be reflective just looking at their reliability scores see this organization has a great deal of personal meaning for me I do not like the part of family in my org oh I do not feel like part of family and my organization that's reversed obviously you caught that good I really feel as if the organization's problems are my own yeah so these are all reflective good look at these the organization deserves my lord loyalty I would not leave my organization right now get more loyalty stuff and these ones it would be very hard for me to leave my organization right now would all right that feels a lot like NC one of the major reasons I continue to work for this organization okay so there's changing costs transition costs but boy those are very similar I'm guessing CC and NC are gonna load really strongly together so let's see what happens factor analysis let's go ahead and throw these all in here and descriptives I'll do a camo and reproduced as usual extraction I'm gonna leave it as principal components for now because we do expect those factors to be highly correlated options I'm gonna suppress more coefficients I'm also gonna sort by size well not to start let's let's leave it off to start and I didn't suppress properly let me go back in and suppress one more time at point three I just cleaned it up a little bit easier to read them okay my was good no major concerns here in the extraction column it extracted four factors which is not what we expected we expected three so we go down the pattern manger to see what happened what a mess we see that poof the only ones loading highly on AC aside from this one NC um ac six and five that reversed maybe AC to MC six is a problem now we say the CC and NC would probably load together you can see there are some loadings together here and see strongest loadings are these top two otherwise they're all over the place and then CC is not great anywhere so what could we do we could take a really fierce approach just right off the bat and try to reduce this to a couple items per factor which isn't ideal but it might end up with the most valid factors so if I were to do this just right off the bat I'd say let's keep a c5 and 6 NC 1 & 2 & CC maybe five and one let's just let's pretend that that's what we're gonna do real quick let's just see what happens so again that was a see five and six we're gonna keep so drop everything else again this is a this is an extreme approach I just want to see what happens and see one and two is what we're keeping we're gonna get rid of the recipe on C and then C C we were talking about maybe one and you know I'm gonna do one two and five vote for CC get rid of three four and six and let's just see what happens terrible not terrible you know middling kmo nothing terrible here we are explaining 60% of variance but look we only have two factors extracted come down here to the pattern matrix and and see an AC loaded together if we were to force this out the three let's just try that real quick forced out the three boom you okay and we end up with seventy-two percent of the variance explained pattern matrix and they do load separately look at that so if nothing else works there's your solution right there back here if we were take a softer approach let's go back to our factor analysis stick everything back in its to give it out first and then everything back in an extraction totally based on eigen values right now okay then we go back down to the pattern matrix here it is and we take a softer approach first thing we do is get rid of all the hardest stuff the cross loadings so MC six clearly a problem it loads so strongly with the ACS so that's the first one I'm going to drop and see six take that out okay back to pattern matrix and we have three factors now not clean factors though yikes for AC mess a mess AC and NC you're loading together very strongly which is surprising I thought to be MC and CeCe okay so what do we do we take I would take off the strongest loadings from AC that are on NC so in or you know let me take off my for because it's the one loading strongest cross loading so NC 4 you're gone and c4 is right here you're out of the family next pattern matrix here we go that looks a little bit better ooh strongest cross loading right now is CC 5 I'm gonna take that guy out CC 5 to make this bigger there we go CC 5 you're out next the pattern matrix and next strongest cross loading is CC - yeah CC - so your app I can Alice's CC to add a matrix you notice I'm just doing this based on the strongest cross loadings oh that's a problem alright but we only have three items oh four items CC one is a big problem how did I not see that before yikes CC 1 CC 1 you're gone better matrix here we go that changed things up a bit ac 1 & 4 are a problem so let's drop both of those together you see 1 & 4 typically I wouldn't do that but I'm short on time I have a meeting I have to go to so look in bed early the a/c is pretty good and C not terrible and then CC wow we came up with something cool

Channel: James Gaskin

Views: 26,529

Rating: 4.821229 out of 5

Keywords: EFA, Factor Analysis, SPSS, Statistics, SEM

Id: oeoTpXiSncc

Length: 8min 7sec (487 seconds)

Published: Sun Oct 16 2016