Confirmatory Factor Analysis (CFA) with AMOS - Example 1

we're going to run a confirmatory factor analysis using the same data set that we use for the exploratory factor analysis video tutorial and that's up here in the efa section here's the data set and i'm simply going to rename it confirmatory factor analysis but there's the video on how we got here and here are the items that list under each component so this is what we're going to go by to run our confirmatory factor analysis let's go so once you've opened up the file in spss go ahead and go to analyze we're going to open up amos step one get the data so i'm going to go ahead and open up the file file name keep mine on my desktop cfa right there important don't have any missing data amos doesn't like missing data so make sure none of the data is missing click ok so now we have our data set loaded up all right we got one two three latent variables let's go ahead and build them first thing you're going to do is click this icon now be careful about making this too big especially if you have a lot of items underneath one new factor it could get difficult and it won't show up very well on the screen so don't make these first latent variables too big and let's count how many of these we have for the first one one two three four five six seven for the first one so we go one two three four five six seven the next one had eight i believe one two three four five six seven eight yep and the last one had three so this one's gonna be eight one two three four five six seven eight and the last one had three one two three now we're going to move these around a little bit and we use our move them around this rotates them and i like mine on the left and we're going to use our little fire truck guy here to move them around pretty actually i'm going to stretch these out a little bit so when the numbers show up for these correlations they'll be able to see and let me move these about you gotta hit this little red fire truck now amos does funny things sometimes and just you know breakfast makes perfect it's like playing a violin so we're gonna move them here and move these guys here and move these guys here i like to keep them in a row there so they're nice and neat what i'd like to do next is to go ahead and name these guys so if you double click on that on the new latent variable it brings up objects properties box and just name it and since we're not sure what they're measuring we're just going to call the first one first you click on the second one and we'll just name that one second and you click on the third one and we'll call that one third so those are our three latent variables construct variables component whatever you want to call them and very cleverly named first second third and we're gonna co-vary them i like the covariance just by doing this you get the covariance area right there you go i want first to second first to third get back in there first to third second a third so every possible pairwise combination is there it's not very pretty that's why you use this magic wand here and it'll shine it up for you nice all right now we're going to load up each item here so we're going to go to this box so here's all the variables all your items and i have to pull up the sheet give me a second here i forgot which it goes where just draw that over here somewhere all right so [Music] item one has question five sixteen is next 11 6 12 eight and numbers lucky number 13 is the last one okay next one uh under factor two it looks like 17 18 2 10 etc etc let me go ahead and move those over 17 18. to you'll notice real quick you'll notice that two question two is a negative that's the only negative one in there so i'm thinking it probably might have been a like it should have been reverse coded or something that is a very common mistake with researchers they they miss something that they should have recoded and just watch out for that okay but i could tell you right now question two is gonna be a problem and where were we question two question ten fourteen three one last but not least is question four all right that takes care of those guys and then the dinky one down here it's question seven nine and fifteen seven nine and fifteen got it all right next thing we're going to do is we're going to fill in these these little circles these circles are actually error terms it's pretty easy simply go to plugins up here name unobserved variables and that gives you all your error terms all right now we've got to fix the output tab you're going to go to the analysis properties and we're going to start with this one is always a default maximum likelihood we want estimate means and intercepts and i think those are all good numerical don't mess with that bias don't mess with that output yeah you're going to make some changes here and we need standardized estimates residual moments modification indices this is a big e right there that's the one that gives us all the numbers we want and i believe that's about it we don't care about the other stuff and don't forget to save it i suggest you save it frequently and make sure it's in the right folder here hold on a second where did i put you there you are right there we'll just call this cfa video aaa i like to keep multiple copies there and here comes the magic time hope it works right a lot of times if there's anything wrong with it this is where the the metal hits the pedal there you're gonna go ahead and hit the calculate button and whistle happy tune what do you know it looks like it worked here all right you'll notice that these loading factors are all over one it's because they're unstandardized you gotta switch to the standardized estimates and and that these are their loading factors so we're looking at them here not very impressive i'm afraid there are one two that are seven or above there's a .69 that's borderline the rest of them are pretty bad which means that factor one under confirmatory factor analysis is not a good factor even though it was under efa looking at factor number two i don't think it's much better the loading factors are all under 0.7 which again means that even though they showed up together under efa they do not load up very well under cfa so factor number two is not a good factor either let's look at the last one well i'll be this one is a good loader upper you got 0.76.80.77 so all three of these do load up well so the third factor does work well according to cfa all right even though that we have determined that these aren't the greatest of uh new factors because their loadings aren't that strong this one is but the other two are not we're going to go ahead and look in the output and don't be surprised if you see things that say that it's a good fit because there's there's like five or six different indicators that'll tell you if this thing can hold water or not let me get it on the same page for [Music] you all right nice and neat first thing we're going to do is go to the model fit so your semen is your first indicator but i it has been my experience that this thing is always significant this is your chi-squared number and what this is basically saying that there is a significant difference between your model this model we're working on and the what they call the saturated model what i call the perfect model so in other words your model does not fit the perfect model strike one boom and i know it has a lot to do with sample size so but let's keep going here's your fit indices the main ones here that used to be gfi i don't know if this is a new version of gfi because this is version 26 i'll have to look into that and get back to you but these are your fit indices that's what the f i stands for fitbit and you want them to be over 0.9 and they all are so that's pretty darn good there's a couple other things we should check we got to look for the rim c and this should be under 0.05 it's pretty darn close but this one right here sinks your boat this p close should be over .05 i'll say that again this p close should be greater than .05 and the rim c should be under 0.05 so there's there's two three more strikes you're kind of out so this p close let me try to explain a little bit better p close what this is testing is a null hypothesis that the rim c is not greater than .05 say that again the p close is testing the null hypothesis that states that the rim c is not greater than .05 which means this you reject that null right because the p-value is less than 0.05 here which means that rim c is significantly greater than .05 and again that syncs your model but we're gonna we're gonna check a couple other things real quick we should have looked at the discriminant validity and we could do that by looking at the picture and we look at these these covariant numbers here they should be it should be less than 0.5 roughly but let's see if i can get a better picture of them let me see if i can't scroll up here i missed the magnifying glass my bad so click we want this bigger we want this bigger so magnify magnify scoot over a little bit magnify magnify scoot over a bit all right so it looks to me like from third to second is .67 too high from first to second see if i can't get this over there a little bit hold on please hold there she be it's .83 okay so that's way too strongly correlated between the first and the second one and between first and third one this looks like an upside down .65 so again your covariances are way too high so you failed to prove discriminate validity there's really no use going on anymore it doesn't matter if we do convergent validity or not but i'm gonna make it official i'm gonna say that this model does not hold water and again for the last time even though it came out under efa exploratory factor analysis in this grouping method it doesn't fit okay it's not a good fit and that is why if you're gonna get an article published you have to use cfa because the publishers will insist on using cfa all right so that's it i hope it helps mgz out

Channel: Math Guy Zero

Views: 6,832

Rating: 4.878788 out of 5

Keywords: Math Guy Zero, Math, Stats, Statistics

Id: DmnINhxuSos

Length: 13min 23sec (803 seconds)

Published: Wed Aug 12 2020