Exploratory Factor Analysis (conceptual)

In this video lecture I explain what an exporatory factor analysis does, and how it works, and why we do it.

[@Gaskination]

Here's a fun pet project I've been working on: udreamed.com/. It is a dream analytics app. Here is the KZbin channel where we post a new video almost three times per week: kzbin.info/door/iujxblFduQz8V4xHjMzyzQ Also available on iOS: apps.apple.com/us/app/udreamed/id1054428074 And Android: play.google.com/store/apps/details?id=com.unconsciouscognitioninc.unconsciouscognition&hl=en Check it out! Thanks!

[@rashidabibi6743]

is this app is free

[@Gaskination]

@@rashidabibi6743 yes

[@쥴스-n3e]

The more I watch your videos, the more I get convinced that you are a genius in teaching. It can't get any clearer and easier to understand the concept of the topic than this, especially with the creative visual aids. Thank you so much for the effort in helping so many.

[@embids]

It's always useful to hear the same content presented a different way. Thanks James!

[@patwheeldon5343]

Wow! Fantastic clear explanation! Oh, thank you so very much! This has been beyond awesome for the breaking down of a complex construct, to enable me to understand. I am so very grateful! from Australia

[@wafajaouadi6566]

Please who can explain me why we use ACP?

[@truegrit7697]

Dear James - thank you so much for this CLEAR explanation! This is truly one of the best videos I have ever seen on EFA. Thanks for bringing this down to earth for us mere mortals! If you have a shrine, tell me where it is - I'll leave an offering!

[@gilessb]

This is the clearest explanation of factor analysis I have yet encountered - thank you.

[@tamammohamad922]

I believe this is the best explanation I have viewed after going through dozens of them. Salute to you and your fascinating way of making it reach smoothly and directly. Cannot wait to follow up with others. You are a great teacher!

[@joybeesoon9630]

Agree👍

[@abhishekasthana1911]

One of the best videos watch on Statistics subject. Thank you James.

[@arashsadeghy1219]

Thank you so much James. I watched this video for the 10th time and I still learned a lot from it. You are doing a great job and I do appreciate your efforts. BTW, you are an awesome teacher. :-)

[@marjiesworld1200]

this was incredibly helpful. you are an amazing educator!

[@tohst88]

You have included the concept for factor analysis. Indeed a very helpful learning video. Thank you.

[@ricardoveiga007]

Thank you very much, Dr. Gaskin, for your helpful videos and statistical tools! I use them and share with my students. Take care!

[@simoneguedes4795]

Very didactic! Great explanation! Thank you, James, for this excellent video!

[@SinghIsKing138]

This explanation is brilliant! Thank you so much for this!

[@carlosm.coaquiratuco5089]

Muchas Gracias, muy buena explicación del Análisis Factorial. Un abrazo.

[@dr.latashawoods4222]

Great visuals with simple explanations!

[@cwmims]

Excellent job James! Thank you!

[@1chelp]

Oops belay that request on the transcript. Just discovered how to switch on the subtitles and will get it from there...thank you again for explaining this topic in such an accessible way:)

[@MrDevid1975]

Dear James, you are great. I really enjoy your videos.

[@joybeesoon9630]

Really good! Clarified a lot!

[@m.c.4458]

Best!!! explanation and animation. Thanks

[@mariaj4239]

This has be super helpful! Much appreciated.

[@darwindhasan]

Dear professor Gaskin, thanks for the good explanation. Like to understand the concept further clearer. Here is my question. Let's say, if we rate the questionnaire items (100 samples) as seen below this way. Will that form as an expected constructs? Construct 1 Item 1: 7 (Excellent) Item 2: 7 (Excellent) Item 3: 7 (Excellent) Construct 2 Item 4: score 6 (Somewhat excellent) Item 5: score 6 (Somewhat excellent) Item 6: score 6 (Somewhat excellent) Construct 3 Item 7: score 5 (Neutral) Item 8: score 5 (Neutral) Item 9: score 5 (Neutral) Like to understand the movement of these ratings and formation of a construct based on respondents ratings pattern. And what if for construct 1 the item 1 rating = 7, item 2 rating =6, item 3 rating = 7. Thanks.

[@Gaskination]

It depends on how everyone in the dataset responded. But, to simplify, yes, SPSS is looking for patterns of responses that are consistent (although they don't need to be perfectly consistent).

[@darwindhasan]

@@Gaskination Thank you for the reply.

[@doctordrone5692]

So rare to find such clarity.

[@kenanmahmutovic8142]

Dear James, thank you for this great video presentation and easy to understand explanation on EFA. This was very helpful for me in doing some EFA for my PhD thesis. Can you provide a literature reference for rules presented on slides 13, 14, 24, 25, 26 and 27. Thank you very much.

[@dr.modevast1953]

Thank you very much Dr.James,this is really help on my research.

[@1standlasttogo]

great efforts and are indeed appreciated.....cheers etc Dawoud

[@ivelaye]

Excellence explanation and useful information- thanks!

[@drvenkateshkarthikeyan5704]

Wonderful video

[@javiergr190]

fantastic explanation

[@Nellymtz]

Wow! I almost fall in love . Thank you very much

[@fliff0504]

Great video!

[@EagleSlightlyBetter]

Great vid. Thanks. Regarding factor retention, you might also recommend Parallel Analysis, (TLDR: each factor is retained if it accounts for more variance than a corresponding randomly generated factor). See Hayton et al. (2004) "Factor retention decisions in EFA"

[@Gaskination]

I'll have to look into that. I've never played with parallel analysis, and I'm not sure how to conduct it in SPSS.

[@MrSwordweilder]

Thanks for your effort in this; very helpful.

[@YouUndeground]

Thank your for this excellent explanation!

[@ibrahimmkheimer5311]

Helpful video. ..many thanks doc

[@neo-qn1vs]

Very clearly explained......thanks...👌

[@francisarthur2983]

Thank you Professor Gaskin for the powerful and insightful presentation. Please, can I get the slides on this video? Thank you

[@Gaskination]

The slides are available here: statwiki.gaskination.com/index.php?title=EFA

[@francisarthur2983]

@@Gaskination Thank you Prof. I am most grateful

[@zuleyhaylmaz1999]

This is very helpful really thank you..

[@SalihFCanpolat]

Doctor, if possible I'd like to access to the slides given at the end of this video. Thank you for this wonderful lecture, you've filled the gaps I've had about factor analysis.

[@Gaskination]

email me for the slides: [e-mail]

[@nitinkumar7225]

Amazing ! Thank you for this video.

[@marieisabelletrimidal9315]

Thank you! This was very helpful!

[@aymanzein7]

Thanks the video.. Informative

[@curlyprivat22]

"pretty fancy statistical sweetness" xD I really liked that phrase.

[@ariskonstantinidis9475]

Great visualization!

[@ebnouseyid5518]

Thanks Dear professor super video. I have a question: The assumption in EFA, that the measurement errors are not correlated it is valid in reality?

[@Gaskination]

I would guess that errors are correlated in reality. We can actually test this using the non-redundant residuals matrix (check the box for "reproduced matrix")

[@ebnouseyid5518]

@@Gaskination Thanks, How I can test ? . That’s what I haven’t been able to do yet. Thanks for givinging me an example of this

[@Gaskination]

@@ebnouseyid5518 Like I said, check the box for reproduced matrix in the EFA window (I think it is when you click on the descriptives button in the EFA window).

[@paparinayak5039]

Respected Sir, I have confusion regarding factor analysis and hypothesis testing. Suppose an exploratory factor analysis extracted 3 factors (factor 'a', factor 'b' and factor 'c') and combining these factors (factor 'a', factor 'b' and factor 'c') we find a construct 'Z'. If we see the relationship of factor 'a', factor 'b', factor 'c' and construct 'Z' with demographic variables like age then we find (sometimes) age has a significant relationship with factor 'a', factor 'b', factor 'c' and age has not a significant relationship with construct 'Z'. If age has a significant relationship with factor 'a', factor 'b', factor 'c' then why not significant relationship with construct 'Z'. Thank you in advance.

[@Gaskination]

I'm assuming construct z is a 2nd order factor. In this case, there will be more error associated with construct z (because of a greater distance to some central point for all factors) than for the individual factors. The more error there is, the lower the critical value, and the higher the p-value.

[@paparinayak5039]

@@Gaskination Thank you very much Sir for the reply. Central point for all the factors means the mid point of a bell curve. I would be grateful to you if you explain little bit more. Thank you so much Sir.

[@Gaskination]

@@paparinayak5039 I mean something more like a factor centroid, rather than a central point on a bell curve. The centroid is the point at which the distance (correlation distance) of all items for that factor produce the least error. Think of the item relationships appearing on a plot. The centroid would be the point that reduces the total euclidean distance to all items.

[@therealjakub]

Beautiful

[@joybeesoon9630]

Do u have for CFA? Thx a lot James.

[@Gaskination]

Here are all my videos on CFA: kzbin.infosearch?query=cfa

[@aparnasm9040]

Dear Sir, First of all, thank you for the wonderful videos. They are really helpful. For my study, I have collected data at the individual level, but I need to convert that data into a group level. I got to know that we can do that through aggregation. But I have a doubt, like when to do it before EFA? OR after EFA or CFA? Kindly help.

[@Gaskination]

You can do it before EFA.

[@sofiaguardado9650]

Hi! Thank you for answering my previous questions. I have another doubt: How should I interpret statements measuring different things (but similar) for example Goals and Intentions, loading as one unique factor?

[@Gaskination]

They are probably two dimensions of a higher order factor.

[@1chelp]

Hi Dr James would it be possible to get the transcript for this lecture? You explained this so well and i understand it so much better now...plus its prompted me to ask other questions in a different way about the topic and encouraged me to find the answers. Thank you very much.

[@Gaskination]

Right next to the Share button, there is a ...More button. Click on that, and then select the transcript, and then the language. That will show you the automatically generated transcript. Hope this helps.

[@akshatsingh6821]

Perfect video..thx a lot

[@sakhhichhabra5]

sir is there any video where u explain how factor and cluster analysis are different?

[@Gaskination]

I don't have a video dedicated to this. The short answer is that a cluster analysis is the grouping of rows and a factor analysis is a grouping of columns. For a more thorough explanation, see this conversation: chat.openai.com/share/ab7f4505-b4e3-4faa-b28f-157000017fcf

[@rashidabibi6743]

if we adapt a scale as it is without adding and reducting any statement but only translating it. can we go for factor analysis or only check the validity and internal consistency?

[@Gaskination]

You can jump to confirmatory factor analysis and skip exploratory factor analysis.

[@michaelpregenzer2147]

fantastic video. thx a lot

[@ibrahimbashimam9149]

This ist too damn cool. Thanks a lot !

[@apoorvadk9635]

Thank you!

[@anugrewal5693]

Hi Would like to take your opinion If we take different dimensions from different authors and form a questionnaire then can we use efa?

[@Gaskination]

EFA is good for finding and validating groups of related reflective measures. If the authors are not providing responses to all of the measures, then it will be impossible to combine their data for a single EFA.

[@mtajammalyasin]

it was very helpful !!!

[@lea_lad]

Hi! Really nice video! I've got a question concerning slide no. 26: why are factor correlations better when they are under .7?

[@Gaskination]

In an EFA, this implies that they share less than 50% of their variance and are therefore considered to be distinct or discriminant. This is not a strict requirement though.

[@asqualgebreslassie8360]

There are fewer than two cases, at least one of the variables has zero variance, there is only one variable in the analysis, or correlation coefficients could not be computed for all pairs of variables. No further statistics will be computed. i got this notice what does it mean plz?

[@Gaskination]

It means you have less than 2 rows of data or there is a variable with no variance (i.e., same value for all rows). Most likely the latter. Make sure you're not including a variable that is constant across all rows. Variables must vary.

[@tayyabayousafzai7443]

Quite useful info

[@egresadounimag]

The tutorial of drJames are very good

[@ahcl9542]

Any chance for a regular ppt file for this presentation? This has been really helpful! Thanks!

[@Gaskination]

drive.google.com/file/d/0B3T1TGdHG9aEQVJIUDdMaTc5NlE/view?usp=sharing

[@ahcl9542]

Thanks a lot!

[@bhaskartripathi]

James Gaskin : Spreading knowledge so generously and for free gets you best wishes of thousands of people like me. Wish you the best from heart for being part of solution to problems of the world.

[@dimple4276]

Respect sir, I have query.. While doing factor analysis in SPSS, is it correct to fixed the number of factors as per the already known scale? Thank you in advance 🙏

[@Gaskination]

I use both approaches (fixed and based on eigenvalues). The stronger solution is usually with eigenvalues, but sometimes strongly correlated factors do not separate on their own. So, it is fine to constrain them to a fixed number of factors. EFA is just exploratory anyway. So, the real test of validity will come in the CFA.

[@dimple4276]

@@Gaskination sir thank you so much for the clarification. 🙏🙏🙏

[@sajeeb2005]

informative video.

[@serendipita5823]

Thank you james

[@zoeyxster]

If factor loading is 0 across all rows, does it mean that the variable is a standalone?

[@Gaskination]

Yes. This also means it does not really belong in an EFA, which is for reflective factors.

[@niteshmorajkar2351]

Is factor analysis useful on qualitative data

[@Gaskination]

No. It is a quantitative analysis technique.

[@Simmy56]

Hey James could you share the software used for creating the item bubble plots to illustrate the spread of item responses in the path analytic framework. I found that very helpful for conceptualizing the connection between the theory and raw data.

[@Gaskination]

+Simmy56 I used AMOS.

[@JanoRamos]

Thank you!

[@modernanteros3801]

Thank you!

[@stephenmurata]

Do you still have a blackboard class?

[@Gaskination]

I have a class on Canvas for the executives at CWRU. I am also releasing the Online SEM Course through MyEducator this month. This online course will be available to anyone.

[@fanyang4860]

Just realized that LINCOLN started a streak of presidents with facial hair (beard/and or mustache) and TAFT ended it.

[@Gaskination]

haha. I had to rewatch the video to realize what you were talking about...

[@imdad995]

Dear sir How to find the value for more than two variable i am really confuse about that can you help me in this matter

[@TheAISChannel]

I think you are asking about a latent variable score, or factor score. You can generate these in SPSS by checking the box for factor scores in the save menu in the exploratory factor analysis function.

[@alhaja.6032]

Hi Dr, really I have no problem in conducting factor analysis with SPSS 22 but after screening data some variables were removed, some were classified under other facor which really different from reseach framework. what can I utilize this result ? I hope it makes sence

[@Gaskination]

+ALhaj A. If items load more strongly on a different factor, then you can either delete those items. If the items are from existing scales, this is really all you can do. If these are items that you developed, then you have the option of recombining them into a new factor.

[@alhaja.6032]

Thanks Dr. for your fruitful comment

[@MrDevid1975]

What if p-Value

[@Gaskination]

p-value is very strict and usually too sensitive. I would not rely on it too much.

[@T15A20]

Is this video supposed to be without audio?

[@Gaskination]

There's definitely audio. Please check your speakers and volume. Make sure you are not muted. You could be muted on your speakers, or the video could be muted (put mouse over video to see volume control).

[@luaymaani]

Dr.James.. I have a question that confusing me regarding EFA on my study. I want to study ((and compare)) the effect of perceptions for social media (SM) and Traditional media (TM) on some mediated variables. I have 7 items for both (SM&TM) with exact same wording for each item. after conducting EFA , SM1 cross-load on multiple factors while TM1 does not.In addition, TM2 cross-load on multiple factors while SM1 does not... I know i shall remove SM1 from the latent variable SM and TM2 From TM... but this will result in differences in items to measure SM&TM ... My guts tells me to drop SM1 TM1 SM2 TM2 from further analyses to have exactly the same items to measure the perceptions on both IVs . however, im not so sure if this is the right thing to do. I really need your advice.

[@Gaskination]

It depends on how you want to test your theory. If you're just looking at the difference in the average response, then don't worry about an EFA. Just average the responses for each set of items, and then do an ANOVA. If you need to include these in an SEM, then go ahead and drop SM1 TM1 SM2 TM2 in order to be consistent.

[@luaymaani]

Thanks a lot Dr.

[@k.c.4011]

James Gaskin What's the reason to average the responses for each set of items before doing an ANOVA? Can we just use the raw scores to compare the two group means, or even doing a t-test? And what are the SM1 TM1 SM2 and TM2? Can you clarify? Thank you!!

[@Gaskination]

kc kc I might average in order to get the overall score for each construct. Then we only have to compare multiple groups based on a single average, rather than on the several items. I'm not sure what SM1 and TM1 are referring to. I don't find them in this video. However, they appear to be items from a survey, where the construct name is represented by SM or TM.