SEM Series Part 3: Exploratory Factor Analysis

next on our to-do list is the exploratory factor analysis you can see here in the order of operations we've just done data screening and case screening and all that and now we're on EFA so let's do it we're already ready the data is all prepped for us so analyze let's see dimension reduction factor what we want all of it except ID didn't tree sort what a bugger let's do that first variable view sort by name sure sort ascending nope we're good all right and then I'm just going to move some stuff out of the way that's the age education experience frequency gender and ID I'm going to move all these down to the bottom there we go okay now let's go to factor analysis analyze dimension reduction vector and control a except the bottom hold shift click the last one and stick it over there descriptives I would like to see a reproduced and a kmo I like some maximum likelihood and I'm going to do it based on I ghen values this just lets SPSS decide how many factors to extract now we know we need what is it let's go look at our model we know we need here it is one two three four five six seven variables so we could just constrain it right here fix number factors to seven but let's just see if it plays out all right continue rotation I would like Promax continue options I'm going to sort by size and suppress small coefficients how small I'm going to press all the way up to 0.3 why because I don't want to accept any loading that's less than 0.5 so it's that's point two away which acceptable and hit okay and hope for the best I actually haven't done this yet so we're learning together okay we got a point nine three three four our camo that is good that is very good actually anything about point seven is fine about point eight is good and above point nine is great so you'd if against should be significant it almost always is if it isn't that is a huge red flag that there's something wrong if you look at our extraction column of the communalities table and we're looking for any values that might be below 0.3 if it's below point three then there might be some issues with correlating with other items so everything's looking pretty good there's a point 308 that's borderline decision quality eight do look do the gun and it looks pretty good it's just decision quality eight now I wonder it was decision quality eight one of our problem Child's over here this just in quality one and no not one of the ones with issues I guess it would have just been in this actually not missing the area okay so we're good on commonalities total variance explained we're going to look in this cumulative percent column and we want more than sixty percent so we're good here and look we actually extracted seven factors just like we expected so that's pretty good we're going to skip the factor matrix look at the goodness of fit test almost never will you see a non significant value here but that is what you want something non significant that's alright we have a huge data set three hundred and what was it eighty responses so we're not going to get a significant there because is based on the chi-square which is dependent upon the sample size and it's inflated with sample size so scroll down we look at the bottom of this huge matrix and we see that we have thirty three or three percent non-redundant residuals and that's good anything less than five percent is just fine here's the the one we're looking for the pattern matrix and it's looking pretty good actually oh here we are info qual has some potential issues here all right so what do we do first with this pattern matrix and by the way we don't need to look any lower at this point we need to resolve the pattern matrix hmm well we can do a number of things we can get rid of the lowest loading items like this decision quality fix or we can get rid of cross loading items that's probably a better way to start or we can look for items that don't load anywhere I don't think there are any of those looks like everything loads somewhere so that's good nothing is a negative loading so that's good so let's start this way there is no right answer and so I'm just going to sort of finagle my way through this we see the info acquisition is loading on decision quality so what do I want to do well info acquisition only has five items decision quality has eight items let's start with decision quality because maybe if I get rid of decision quality six or maybe eight then we'll see the info acquisition is no longer very related I don't know if that makes sense to you makes sense to me we learned that decision acquisition eight or quality eight we had a low communality you may recall from up here or is that point 308 there is a decision quality eight so I'm actually going to start with that one even though decision quality six has the lower loading so let's do this factor analysis go to decision quality eight I'm going to drop it out hit OK I'm going to go just straight back down to that pattern matrix we can see decision quality six dropped quite a bit so we can go ahead and get rid of that these still too loose these two still load there so let's just go get rid of number six it is decision quality six okay jump to pattern matrix that's looking pretty good that's usefulness let's go to decision quality it moved down we got some point five s here which they're fine for now we might remove them later we'll see and oh look hey this one in pack five is no longer loading on decision quality so that's good but it is a point four five four so that's not good so let's do this let's go ahead and get rid of em into Act four I think is our best bet because it's loading in both places although the loading is separated by more than 0.2 and it's loading strong more strongly here II there's a tough decision I think what I'm going to do is drop into act five and then see what happens and then drop into AK four and add five back in if necessary so let's start with into AK info AK five removed it okay jump to pet and matrix still that's it that's now it's within point two let's try again I'm going to stick info act back in a number five and I'm going to go get rid of number four and just see if that makes a difference now we don't have seven whoa now they all loaded on a decision quality that's not good so in fact four was holding us together yikes let's go get rid so go add in two AG four back in and drop five okay so we're back to where we were right here this one's overlapping quite a bit whatever let's see if we we still have several items to play with over here so if we were to drop decision quality seven with a help or decision quality one let's find out because those week there's sort of wiggle room right there go down decision quality seven drop that out eight okay pattern matrix there's decision quality and oh look these bumped up and and and they're no longer loading on decision quality oh that's awesome okay and this is mostly fine we could drop decision quality one if we wanted you know I'm just going to do it keep all these fairly high we have enough of them to to work with decision quality one and hopefully that doesn't ruin anything if so we'll just put it back in here's decision quality we have four of them now they're fairly high a lob of 0.7 and all these are fairly high so we're good excellent oh that's a relief okay we're look at the others low point six seven that's five point seven point seven point five that's a low playfulness one yeah it's you know it's above the point five and averaging out I think we averaged over 0.7 so I'm going to leave it point six point six so we're good actually this is very good so now that we have a pattern matrix that we're happy with let's go back up to the camo looks like it's still great significant good extraction do still good and we're actually explaining more variance now that's 66% here's the goodness-of-fit tests not significant still because we have a huge sample science reproduce matrix we're down at 2% that's excellent we know the pattern matrix is good and so what's the next step here we are iterate until you arrive at a clean pattern matrix check adequacy well that's the kmo and Bartlett's and communalities and total variance explained all which were good so we're good there convergent validity let's go look at that convergent validity is do they load highly on their factor and what is high well it depends on your sample size for our sample size over 300 anything above 0.3 is actually acceptable although I would never accept anything that was less than 0.5 and I want them to average out to above 0.7 if possible so this one definitely have urges out above 0.7 this one yes yes yes yes yes and probably maybe maybe not pretty close though but the loadings are high enough to be convergent you might say discriminant validity well that is are there no cross loadings yep new cross loadings within point two and if you go down here to the factor correlation matrix are there any correlations between the factors that are greater than 0.7 so we look look look I see two point five point six is pretty high that's five and seven I'm guessing that's info info acquisition and decision quality let's go check it five and seven here's five that's decision quality yep and seven is input acquisition so these are highly related but not to the point seven percent are not presenting out to the point seven level which is if you just square it point four nine and that is a percent of correlation so there that would be 49% correlated which is almost half which is just too much in this case it's 0.6 squared which is 0.6 for one time is 0.6 for one we're about we're sharing about 41% of our variance between those two factors that's a lot but not too much to handle so we're good on discriminant validity and then reliability ah the rap reliability is cronbach's alphas so let's do those we go to analyze reliability where are yet oh click quality control nope boom classify nope scale there it is scale reliability and I'm going to start with usefulness we didn't remove anything from usefulness so I just throw all seven in there hit OK and I find that I have a point 943 for my cronbach's alpha what do I want above 0.7 ideally above point six is even okay but I would much prefer about 0.7 so that one's great and excuse me where did it go the liability analysis we do another one of all those playfulness did we remove any I don't think we did we thought about it but we didn't do it Oh point nine one two and you can do this for all of them and you want to make sure that everything is above point seven like I said I'm going to go ahead and do this roll them and then make a table so that you can see it I'm going to pause the recording for now okay I finished I did the composite or the cronbach's alpha for each one and then I went ahead and copy the pattern matrix into a Word document and added a new row for cronbach's alpha and I should probably fold that so it's really obvious I probably pull all of it there we go and so now we know for each factor what the cronbach's alpha was that is our reliability and is that everything yes it is we're done with exploratory factor analysis yes

Channel: James Gaskin

Views: 113,816

Rating: 4.8831463 out of 5

Keywords: Screencast-O-Matic.com, SEM, EFA, Pattern matrix, SPSS, Factor Analysis, Measurement model

Id: X-O-OcJPCe8

Length: 14min 14sec (854 seconds)

Published: Thu May 02 2013