Being able to actually see the numbers and how they are input into SPSS and excel in other videos is much more helpful then simply reading how to apply and use from a textbook. All of the videos on SPSS and excel will be good to go back over again and again in helping me understand the stats and tests used.
This was extremely helpful. I didnt find any other video on you tube on factor analysis that mentions how to remove a variable if its loading on more than one component. I read it in a book, but didnt quite understand it. Your video has cleared it so very well. Thank you very much Todd Grande.
Thorough walk through. There was a lot that I admittedly did not initially understand, so I appreciate the depth of analysis. In all likelihood, I will watch it again in the future.
I actually enjoyed being walked thru this analysis. When I see the process unfold, it helps me better comprehend the concepts that underlie the analysis. That was very helpful, thank you Dr. Grande.
I appreciate Dr. Grande discussing the difference between factor analysis and principal components analysis. At first, I thought these terms were used interchangeably. I personally think that SPSS is confusing in itself, so this may be why I found it challenging following the steps in this video.
Exploratory factor analysis simply tries to determine which items measure which variables with no expectations of number of variables or which items measure which variables. Confirmatory factor analysis requires you to set the number of expected components prior to running analysis to confirm results are as expected. What would help would be to use the actual item names (labels) in the results....might make it clearer for students to understand what is going on. The high loading simply means that item is measuring that variable (or construct). If the loading isn't high enough, the item is not measuring anything which is bad. If it loads on more than one component, then it is measuring more than one thing which is bad too. Also, I never like to accept any loading less than .600. I think .450 is quite low. Also, important point to remind students to remove one bad item at a time an re-run analysis...typically the lowest or item that loads on more than one component. Removing multiple bad items at a time might lead to the removal of good items being confounded by really poor items. Good luck.
@@mahasarwar5513 Possibly both...more that likely with what you are measuring. If an item loads on more than one component/dimension, then it isn't discriminant from measuring other factors or variables. Or it could just be a bad or vague question that needs to be removed from the model.
This is a marvelous video for those who have spent the whole morning struggling to understand PCA! I would like to ask what is the minimum value you consider acceptable in the Rotated Component Matrix
Well well well. I watch your new videos for entertainment and when I’m avoiding university work. But here I am at 10:30pm actually trying to do university work and I’m again at your channel.
This video contained a lot of new terms and information to me. It was slightly confusing as factor analysis and principle components analysis are different but are used interchangeably. I will need to watch this again and have further reading on these terms to get a grasp on factor analysis.
Certainly a wonderful video, not very suitable for one with no background about PCA. I am a beginner, and had read three chapters about PCA from 3 different books to get the background. This video showed me how to implement the theory. Toward the end of the video, when Todd does some iterations to find the best number of secondary variables and PCs, I was wondering how many times the process should be iterated, and found the scree diagrams of iterations become more and more similar to each other, till you reach a point where no variable is required to be eliminated form the list. In other words all of them contribute fairly well to one of the PCs. I should say, I am using MATLAB for this. Thanks Again Todd :)
+Melissa Clendaniel I agree this video was confusing. I am no familiar with factor analysis and without using this on a daily basis I would still need to rely on refreshers.
The video is helpful in understanding what a factor analysis does, which is to explore the underlying constructs in an assessment in order to determine what the assessment measures. The demonstration on how to conduct a factor analysis is helpful, but I find it somewhat difficult to follow and would likely need further clarification and practice.
The video was a little loaded for me to follow, but Dr. Grande does a great job of thoroughly explaining the importance of a factor analysis and how to conduct such.
+Candace Fernandez I don't like SPSS so this may be why it was confusing and hard for me to follow. But overall, I also liked the explanation of the importance of a factor analysis.
Could you explain why you mentioned that the correlation matrix value should stay withing -0.8 - (+0.8) because I am getting values such as 0.9 and more. So should I remove those items and try again? What exactly does a value over 0.8 mean?
Hi Dr Grande. Thank you for this video. One question for you, what are we supposed to do if the correlation matrix demonstrates variables with more than 0.8 value? If removal is required, is it the variable in the column or in the row?
I had a hard time following this but did jot down a few statements. Factor analysis will not tell us what the constructs are (that's somewhat of a subjective judgment), but it will give groupings to look at. Principal Components Analysis is actually different that Factor Analysis although they're often used interchangeably. KMO and Bartlett's Test - KMO cutoff rate is 0.5 (ideally closer to 0.7 or 0.8... anything less is starting to cut it pretty close).
very nice video explanation. i have one doubt, after caliculation factor analysis, gouping factors , how to find or compare influence factor with dependent or demographic variable
Sir, could you please explain how to name the extracted factors if two types of variables are loading good on same factor?On what basis the variables are deleted and PCA is run again?
Although this seems very complex I could see how it would be helpful in deciding which items to use especailly when creating an instrument. I can imgine this is what my professors in undergrad did to check to see if a test question was a good question or not.
+Mandy Moore I was pretty lost throughout this video, but I agree, determining whether or not test questions are good questions or not is definitely important and I can see how this method can help to decide that.
what you are explaining regarding the correlation matrix around minute 7 is only applicable to the varimax rotation where you wouldn't expect the factors to be correlated, right? i am using direct oblimin as i expect my factors to be somewhat correlated and i'm wondering whether i should be interpreting that table in the same way as you're describing in this video. any help from anyone would be greatly appreciated!
ohh wait I search for PCA and start listening to his video and im thinking this voice sounds familiar only to check the channel name and it is non other than Dr Grande!!!
can somebody tell me why it would be 'a red flag' if the correlation matrix shows values higher than .8 or lower than -.8? If the items measure the same construct I would think it is normal that the responses are highly correlated. I ran the analysis with my data and it showed me high correlations of sometimes above .8, so what does this change about the interpretation now? I would be SO grateful if somebody can help!
To be honest, I didn't understand a word of this at all. I was confused on how to do it and what the end goal means. I would like Dr. Grande to go over this in class.
Great video Sir, thank you. Could you clarify a little bit on components loading. We look for the item set for which the components "load well" with particular items in rotated component matrix. Would you elaborate more on "loading well"? e.g. 14:19 : why component 4 does not load well with items 7 while 4 having values .844 and .476? both values should be greater than 0.5? or both should be either greater than 0.5 and should similar values? Thank you.
Hello Dr! Can we apply PCA to two variables in order to find an index which would describe both those variables? For example, the first principal component of PCA applied to height and weight as an index of overweightness. If the case is ranking, does PCA have meaning at all or will the result be the same as an average of height and weight? Thanks!
A little more difficult to understand, but I do understand that it seeks underlying unobservable variables that are reflected in the observed variables.
+sherrie tilghman I was very lost on this video. I think it was because there were so many different things to check once the factor analysis was completed. I will need a lot of assistance understanding this.
Hello, Respected Sir, EFA of 79 line items scale shows their are 15 factors but all values are loaded in first column. i dont know why.... it means there is only one dimension of scale ? but i does not seem logical. kindly guide me how to handle this.... thanks in anticipation
Hi..thanks for your video - it will definitley help me in understanding and conducting the factor analysis. I have a few questions and hope you can help me. I am asking questions with respect to doing factor analysis for likhert scaled questions. 1. Is a component formed of less than 3 variables acceptable? will it hold onto the reliability test? 2. Do we need to make the questions unidirectional ie either all positive or all negative? 3. what level of variance explained is acceptable like how low is accepted?
+Deepa Navin For latent variables (constructs), 2 items typically minimal so you can run a reliability assessment. No, simply ignore the +/- sign in the loadings..they are really irrelevant it interpretation. You can re-code the reverse items prior to running a reliability assessment so the alpha is correct and all correlation values are positive. No real answer...but higher than 50-60% is good.
I think its more to see i items are actually measuareing the contruct they are intended to measure. This was complex so I'm not sure i fully understood it either.
There was a lot of information to sift through and to examine. I felt lost for the majority of the time. I will need an additional explanation to understand this video fully.
Hello Dr Grande. I am doing a thesis conducting factor analysis. I am using 3 items in my categories. So I am getting only one component and spss is not running my rotated component matrix. Please help me
Sir , I'm using stata, are there any specific commands for principal component analysis PCA in PANEL DATA Or Just simply run PCA after standardizing variables?
Thanks for the video! Anyway, I want to make sure... If there is any component that performed below 0.45 in Rotated Component Matrix, we have to remove it and check again until all components are greater than 0.45, right?
What happens if I change the eigen value from 1 to 0.5 ....by putting it 1 I'm not getting my rotation matrix...it says only one factor is extracted. Please help
What it means is that there are no variables [items] that can be reduced. Thus, it is better to use the whole items to represent the construct you are intending to measure.
It's confusing that the principal components analysis is considered a factor analysis but isn't? That's what I think I gathered. Because this is SPSS, I found it simpler and the fact that there is a function called "principle component." It is always harder when there are so many different names.
If the data is binomial (1=correct response, 0=incorrect response) then this technique does not really work. However, there is a way to get 'factors' using the R program. See: it.unt.edu/sites/default/files/binaryfa_l_jds_sep2014_0.pdf
I could understand what you were doing when doing the analysis but it went a bit too fast in the actual steps especially when you were checking off the boxes to run the analysis. I will need to watch this agin.