Thank you James for keeping this video here. I haven't done an EFA in a while and your video helped me to remember how to and why. I also had to refresh my CFA-SEM skills and again your videos helped me through the process. Many thanks
Hi James, when doing EFA, after eliminating cross loading in the pattern matrix, there are still 2 items which have loading more than 1, should I delete them as well? Thanks
@James Gaskin Hi James, what should I conclude if the loading is more than 1? Can I just say that the variable is strongly correlated with itself? Thanks again.
James Gaskin thanks, i have been watching your videos for about 5 hours today to find out the answer :D but thanks anyway. I really like your channel. It saves my life :D
Hello James, I have a very urgent question. (If anyone else can explain would be great as well!) Why do you do EFA If you already have the theoretical constructs? In the Book of hair et al “multivariate data analysis” it states, that you only do EFA when you don’t know anything about the structure of your indicators. But if you know the structure due to theoretical backgrounds, you only do CFA to confirm these structures... I am so confused and would be so greateful if Someone could explain! Thank you!!
A non-trivial amount of debate exists among methodologists regarding whether the EFA is absolutely necessary, particularly when the set of observed measures have all been validated in prior literature (Costello and Osborne 2005), and when the factor structure is already theorized (i.e,. when we already expect, for example, that the jsat1-5 variables should factor together to represent the job satisfaction construct). As has been shown in replication studies (cf. Luttrell et al. 2017; Taylor et al. 2003), the same scales will perform differently when in the presence of other measures (not in the original study) or in the context of a different sample. Thus, in my view, the EFA should always be conducted to surface validity issues idiosyncratic to the current context. I will not resolve this debate in a footnote. My personal school of thought is that it is best to do an EFA first because discriminant validity problems manifest more visibly in the EFA than the CFA. Then follow this up with a CFA informed by the EFA. The EFA is for exploration only, and should be used mainly to highlight potential problems (such as discriminant validity) that will likely resurface in the CFA.
Hi, Professor Gaskin. Thanks for this amazing series! In this video, you mentioned something about mitigation strategies (min. 4:30) I couldn't find any specific reference in your list of videos, although I still have many more to go through. Could you please indicate some specific references on mitigation strategies? Thank you and kudos for this channel !
Professor +James Gaskin , thank you! I"ve already found some tips and tricks embedded in many videos, including this one, but I felt the need for something more specific. Again, I must express my gratitude for your videos, they're some of the most useful resources on CBSEM (along with your work on PLS) that can be found on the Internet. Congrats!
Dear Pr. Gaskin, thank you for the video. I have the following question: the dataset I am running the analysis on has many 996(not sure what to answer) answers, and those are spread all over the dataset, which makes it complicated to run the analysis. Can I run multiple imputation or EM imputation to replace them with something more sound or is it falsification of the results(since technically they are not missing values) and adding more bias to the data? If not, I don't know what can safe the analysis, because if I use listwise/pairwise deletion, I end up losing more than 2/3 of the dataset...
That is too much data to impute unfortunately. I try to avoid including n/a or unsure options unless they are needed to prevent invalid responses. You can try running without those values, but if it is too little of a sample size, then you might have to run smaller EFAs (with fewer items).
Hello Dr. Gaskin. I hope you find time to clarify the following doubt. While doing EFA as suggested I noticed that two theoretically distinct factors were loading as a single factor. When I had run an EFA on those two separately, without any item removal, they had identified as two different factors but when added to the remaining variables for an EFA, these two again load as a single factor. I understand now that I would have to do my EFA for these two separately and consider them as second order constructs and another EFA for the rest of the variables in the model. My doubt now is that should I still stick on to the threshold requirements of total variance explained (
With only two factors, the variance explained might be low. That is understandable. Same issue with the non-redundant residuals. Fewer items means each residual has more influence. So, in both cases you can be more lenient.
Hi James Thanks for the wonderful work My question is Am i supposed to delete some factors in case the EFA comes out with less factors (imagine in this case we got 5 factors instead of the 6 factors that were being expected) Regards
Try to keep all theorized factors. You can even constrain the factor analysis to extract exactly the number you want. See if you can make that work. Also, make sure all your factors are reflective and also that they don't belong to a second order factor.
Hi James, thank you so much for these videos. They have been so incredibly helpful. Towards the end of the video you state that some literature makes a case for a Chronbach's alpha of .6 (rather than .7) especially if you have around 2-3 items. Do you by chance have a reference for this?
Hi James, many thanks for this video. I have a complex model that consists of dependent and independent variables that are multidimensional. Can we use the same EFA method in this case? Or should we do an EFA separately for each variable?
Hi Professor +James Gaksin, thanks for the fantastic videos. Really appreciate your time and effort. I have a general question concerning reliablity: is it possible to use Cronbach's Alpha as a reliability measure in a multi-group case? I expect a large amount of variation between the groups which would distort the Alpha, would it not? I would appreciate your feedback. Thanks and kind regards!
Usually multigroup effects are theorized at the structural level (relationships between constructs), rather than at the measurement level. In fact, to even test multigroup effects at the structural level, you must demonstrate no differences at the measurement level (invariance). So, cronbach's alpha should be stable even when there are multiple groups, unless you expect these multiple groups to be different at the measurement level.
Dear James, Thanks for your awesome work here. I already learns many new things! However, I still got some Questions. For my master thesis, I want to find predictors for quality of life. To measure the quality of life (QOL) I used an already existing questionnaire with 21 Items, which are represented in 2 Dimension of QOL (eg. Mental, Physical). Do I still have to run an EFA for that questionnaire? I expected 6 Factors and got 11 when I run an EFA with all my variables. Many variables are loading in 2 factors. For example with 0,3 in 1 and 0,7 in 2. What can I do about it? Some others with -0.3 in 1 and with a value of 0,7 in another factor. Best regardand thank you.
I always run the EFA, but sometimes you can get away without running it if you are using only established scales. With multidimensional constructs, it is sometimes best to run the EFA separately for each dimension. So, if you expected 6, that means you probably expected 3 for mental and three for physical. So, run those two group separately. Crossloadings are fine as long as they are sufficiently distant from the primary loading (I use 0.200 as my difference threshold, but others use 0.100).
Hi James, thank you so much for these tutorials. You are walking me out hand by hand! I have a question on the EFA. There are 12 items and the KMO is .899, Barlette 983.25, df 66, sig. the covariance of all 12 items are bigger than .3; but i loaded only 1 factor, accumulating 59.841%. Why I have found only one factor? is this suit for EFA? Thank you!
This means there is a strong correlation between all of the items. Try different factor methods (ML, PCA, PAF) and rotation methods (Promax, Direct Oblimin, Varimax), and try forcing it out to as many factors you expect. If none of this works, then consider whether method bias might be strongly influencing your data.
@@Gaskination Thank you. The instrument is from literature but I didn't see any EFA or CFA on the instrument. Can I force EFA or just do CFA? Using CFA with 12 items under one latent is acceptable? Thank you,James. I'm following your STM series step by step. Words can't express my appreciation to your help!
@@xiaoyuan1680 There is debate whether an EFA is required. The short answer is that it is not required, but it is helpful. You can constrain/force the number of factors to see if it works. If not, you can jump to a CFA to see if it is any better. Just watch out for discriminant validity issues.
Dear James. Again thanks a lot for this very useful video. In my model, I have 12 constructs(IVs, mediators and DVs). When I put them all together, only 8 factors emerge. If I exclude the very problematic scales (3 of them), I need to fix the number of factors in order to obtain 9 factors otherwise it is 7. So how should I proceed? Any reference for fixing the number of factors? Any references for any other suggestions? maybe you can upload a messy model like that and show how to deal with it.
If any of the constructs in your theory are 2nd order, then you may want to do a separate EFA for those. Otherwise, it will be difficult to argue for 12 factors (or 9). The other option is to skip the EFA for now, and go straight to CFA. Get discriminant validity sorted out during the CFA and then maybe return to EFA.
This happens all the time. You just need to make sure the variable loads most strongly on the factor it is supposed to load on, and that its "cross-loading" is at least 0.100 less than its primary loading (but ideally 0.200 less).
Thanks Mr. Gaskin for ur video. btw I have question, what we have to do if we have non-redundant residual >5%? Actually I ran EFA just like what you did in the video, but I got non-redundant residual=7% My KMO is okay, but my "Extraction Sums of Squared Loadings" is 51%
7% is fine. Some say even up to 50% is fine. Most don't report this measure anyway. The Extraction Sums of Squared Loadings at 51% is fine. It is low, but not too low. Anything less than 50% is discouraged.
Hello professor. While we have 3 variables, and each variable consist 3-5 items. It is necessary to do EFA with all items, so it is going to be reflective on its variable? Or it is a must to do EFA one by one for each variable?
It is best to include all items of all reflective constructs in the EFA. This will then allow you to assess discriminant validty. If tested individually, you cannot assess discriminant validity.
I have a query related to the reliability of items. I have a latent variable that consists on two observable items. These items are taken from past literature. One item is reverse coded in previous papers. When I tried to reverse code it, it is showing the reliability of 0.58 and I tried both ways of coding but the reliability still remains same of 0.58 in both cases. Why?
Hi!! thanks a lot for the video!! A quick question plz: if I have a formative construct with first-order reflective dimensions, can I use a PCA or an EFA in SPSS?
Hi James, how about goodness of fit test which indicates the difference between the estimated and sample correlations, and thus indicates the factor model is a good representation of the relationships amongst the observed variables or not?
During the EFA, you can assess this if you use Maximum Likelihood approach. it will produce a goodness of fit measure and the degrees of freedom that you can use as a denominator. If the quotient is between 2 and 5, then you're in pretty good shape.
Hello James, I have no cross-loadings in the pattern matrix; however, 2 additional factors have been created mainly due to one variable, which is making 3 different factors. What can be the issue and solution for this?
Hi James! Thanks for the informative video. I noticed that in my EFA, one of my variables loaded above 1, is this a problem? I found that when I changed the rotation to Oblimin, it reduced from 1 to .718. Is that a problem? Furthermore, I did have separate loadings without any cross loadings, but unfortunately a variable I theorized (identity affirmation) loading neatly onto my mental health latent variable instead of what I originally theorized (under resilience strategies). As such, is it fair to drop it entirely from my analysis?
The heywood case isn't much of a problem during the EFA, since it is only exploratory. As you note, you can usually hide it by changing the rotation method. When a set of indicators load with another set (unexpectedly), I would recommend running a separate EFA with just those items (from each construct), to see if you can separate them by removing one item at a time. If they cannot be separated, then consider whether you might have a 2nd order factor with multiple dimensions.
Hi James, I have a question to ask. My conceptual framework is unlike yours where 1 observed variable has 1 item. Each of my observed variables have at least 1 items. I do not know how to use EFA in this case. For example my latent variable consists of 3 observed variable each with 2 items.
If a variable does not have more than one indicator/item, then it is not latent, and therefore does not belong in the EFA or CFA. You can bring these single (observed) items in during the causal model.
If you do it separately, then you cannot assess discriminant validity. Discriminant validity assessment is the primary strength of the EFA. So, I would strongly recommend a combined EFA.
Dear James... I am building a questionnaire on videogames as a part of my thesis. I have 8 variables under 4 blocks, each with 4 items: intrinsic motivation towards games (based on a scale by another author), then three varaibles that have to do with player needs (autonomy, competence and relateness) based on a scale by another author and finally a block on importance given to game elements (conventions, components, actions and emotions). Im not doing SEM and EFA and CFA are only to validate the pilot. Should I do factor and confirmatory for each block independently as they are different? Even though It will be easier (avoiding cross loadings, etc.), I think it would be more representative if I do them together at least the first two blocks together so I can be sure I am measuring what I want, isnt it? Big thanks!
Thanks so much! Still, to my understanding, if there are particular scales within the questionnaire (like the one that have to do with player needs), in the confirmatory you want to check that this scale is "three-factorial" and you should run it independently, don't you think?
Hi James, Thank you for your efforts. I have a problem. I am running EFA for a framework that consists of multidimensional constructs. i.e. i have technology factors that include sub-variables such as security, privacy, trust and other multidimensional such as organizational factors which include top management support and organizational readiness. My question: is it right to run EFA separately for the organizational factors and then for the technology factors? if yes, is there any reference to support that? Thank you very much!
It is right to run them separately, but I'm not sure if there is a good reference for this. If you find one, please let me know and I'll put it on the statwiki.
Thank you. I only found indication in Hair et al. (2010, p.101) referring to the fact that it is inappropriate to run the IV and the DV in one EFA. I found also other literature but not really reliable. I think it is a good idea to write paper about it and to compare full EFA with separated EFA.
Hi James, I am using a 33-item measure (Driving Anger Scale) and using promax rotation leaves me with quite a few loadings below .4. Would it be fine to use varimax? It has provided me with stronger factor loadings. Cheers
Hi Mr Gaskin. Good Morning How to perform EFA if DV dont have any sub-dimensions? Can we run EFA for DV along with any other construct? Question.2. Is there any need to perform EFA if construct has only 3 items (with no sub-dimension)?
Definitely collectively. This is so that you can determine the discriminant validity between variables. This is the primary purpose of an EFA (to show factors and how they are distinct). So, doing an EFA with only a single factor undermines the purpose of the EFA. No reference needed since this should be common practice. You can also examine any article using SEM in a top journal. They should follow this approach.
Hello professor Gaskin Thank you for this informative video. Sir one question. Do we need to run EFA on established scale or can i directly go for confirmatory analysis directly??? Sir if you have any reference that justify applying CFA directly without EFA. Please share it with me 🙏 I'll be forever great full to you as I have run CFA without running EFA in my dissertation.
EFA is not necessary if all measures are already validated in prior literature. This is established process, so no citation needed. If you really want to cite someone for it, you could cite Hair et al 2010 (Multivariate data analysis), as there is a whole chapter on EFA.
@@Gaskination my lecturer said "The prior theory validity will be different if you are doing research in different places, then you should go for EFA first" is it true Mr. Gaskin?
@@brantazconflix3873 I agree. Even with validated scales, an EFA can be useful if studying a new context/population. However, strictly speaking, the EFA is primarily used for new scale validation, rather than re-validation of established scales. Most would probably suggest to go straight to the CFA for established scales.
Hi, James, is there a reference you could provide me in regards to your determination for an acceptable amount of non-redundant residuals with absolute values greater than 0.05, and why there is a recommended 5% ceiling? - Thank you!
hmm... I've never really thought about that particular threshold. It is just what I was taught so long ago. My guess is that it comes from Hair et al 2010, but if not, then maybe from one of the ones on this list: statwiki.kolobkreations.com/index.php?title=References#Exploratory_Factor_Analysis
I have found the best way to resolve this type of issue is to do a separate EFA with just the items from the offending factors. Work out this smaller EFA (by removing items one at a time that have the worst cross-loadings), then reinsert the remaining items into the full EFA. This will usually resolve the issue. If it doesn't, then consider whether these two factors are actually just two dimensions or manifestations of some higher order factor. If this is the case, then you might consider doing the EFA for this higher order factor separate from all the items belonging to first order factors. Then during the CFA, make sure to model the higher order factor properly by making a 2nd order latent variable.
Hello Dr. Gaskin, I have a question regarding running an EFA. When I run mine, it says that "Rotation failed to converge in 25 iterations." My KMO and Bartlett's test as well as communalities are acceptable, so I'm wondering what could be happening?
It is most likely either low sample size (compared to number of variables), or there are one or more variables that don't belong (e.g., formative indicators, categorical indicators, grouping variables).
I experienced this. For the same data set, I allowed SPSS to determine the number of factors and came out with 9. I tried to set the factors at 5, I got the same message- rotation failed to converge in 25 iterations. I tried 6 factors and it worked. I surmised that maybe at 5 factors, the %cumulative variance is less than 50% under the rotated variance explained column.
Hi James! Can I give my questionnaire to my experimental and control groups even BEFORE conducting EFA? Due to lack of time to learn and run EFA, the questionnaire I gave to my experimental and control groups is not yet reduced and NO identified factors are determined as of yet. My plan is to conduct EFA later on. When reduced items are already available, that's the time that I will be selecting appropriate responses from my study groups for my analysis. Please comment. Thank you
Thanks for your reply. I asked the question because, initially, I thought EFA should be done first before giving it to my study groups (experimental and control). Anyway, I will be doing EFA later. I will be analyzing (t-test, etc) my study groups' responses BUT I will be removing their responses to some items in the questionnaire that will be excluded by the EFA process.
All factor loadings of one construct are negative in my Pattern Matrix of EFA, does this happen due to the reason that this variable is negatively related to the dependent variable? Factor loadings of all other constructs are positive in the pattern matrix. All other results are fulfilling the criteria
Reliability is estimated in a silo (ignoring all observed measures not in that particular set). EFA accounts for variance shared across all measures included in the EFA. This means that if those particular measures share a lot of variance with other measures, then the amount of variance loading on their home factor may decrease. This is a lot like the difference between correlation coefficients and regression coefficients.
Hi Mr Gaskin, you mentioned about scores, could you please explain a little more about it. How can we use it? if you have a video, kindly share the link.
Factor scores are a weighted average of the items that have loaded onto the factor. These scores are weighted by the indicator weight (or loading) shown in the pattern matrix of the EFA. I don't have a video for it, but the procedure is pretty easy. Once you have a good EFA solution, redo the EFA exactly as it is, but go to the scores button in the EFA menu, and check the box for save scores as variables. When you run the EFA this time, it will create these factor scores as new variables in your dataset. Each factor score will represent the items that loaded on it as their primary factor.
Would you kindly provide a link where you talk about mitigation strategies if the number of factor extracted do not much your model ? (Part 4:30 of the video)
Sorry for the delayed reply. I've been busy trying to fix the StatWiki. It is fixed now though. As for your question, you can check out the 'messy EFA' video: ruclips.net/video/oeoTpXiSncc/видео.html
I have 5 constructs in my model. In used 4 "adopted" measurement scales and 1 "adapted" measurement scale for analysis. Do I need to perform EFA for all or only for the one which I "adapted". any reference article will also help along with your kind suggestions. thanks a lot
It is always best to do the EFA on all reflective latent scales, even if they have been validated previously in the literature. The EFA allows you to detect discriminant validity issues specific to your set of constructs, and convergent validity issues specific to your sample. As for a reference, here are several: statwiki.kolobkreations.com/index.php?title=References#Exploratory_Factor_Analysis
Hello Sir, Can you elaborate about the 'Measure' lile Scale, Ordinal and Nominal. I am working on Theory of Planned Behavior. What do you suggest the 'Measure' to these Psychological Variables, i.e. Scale or Ordinal..... Thanks..Waiting for reply
Usually TPB constructs are measured on ordinal scales. Ordinal is discrete interval scales (such as Likert e.g., Strongly Disagree to Strongly Agree). Continuous is not discrete (such as age, income, years of education, etc.).
Sir, I carried out EFA and got 6 factors as per the model. Kindly guide me on how to interpret the results in APA style. I am searching for complete interpretation of results obtained from EFA, CFA and SEM. Please help me.
Hi Prof James, Is there any video example of doing EFA for higher order model e.g.: 2nd order model? I've tried to do the EFA based on this video but the pattern matrix always show 'rotation failed to converge in 25 iterantions' Please let me know what did I do wrong. Thanks
For models with higher order factors, I usually conduct a separate EFA for the items of the higher order factor (so that just its subdimensions will factor) or I skip EFA (if the measures have been validated already in extent literature), and then just go straight to the CFA. Another approach is to use principle axis factoring instead of PCA or ML.
@@Gaskination Hi Mr. Gaskin, Thank you for your reply. The reason I need to run EFA is to solve a discriminant validity issue of one of the variables as AVE
@@arifnugroho73 You can be less strict on discriminant validity between dimensions of a higher order factor. These dimensions are expected to be highly correlated since they are manifestations of the same thing.
Yes, when sample size is not sufficient to use separate datasets. It is considered more rigorous to use separate, but very few studies actually have sufficiently large samples to accommodate separate analyses.
I have 7 latent variables in the model. I took all the observable items from the literature. Still, there are 2 latent variables that are loading in the same column pattern matrix. 6 factors are being extracted. They should load on different factor columns. What can be the solution? Is it sample size or there any other thing that is causing this issue? Waiting for your reply..
@@Gaskination Why the Pattern Matrix wouldn't able to differentiate the 2 factors in column matrix? Is there a chance that the observed variables are less for that certain latent variable to differentiate among them? Can you elaborate a little bit about it as the variables are already taken from the literature? Thanks in advance.
@@abdulmoeed4661 Here is a post I wrote about it: gaskination.com/forum/discussion/98/what-to-do-if-two-factors-dont-pass-discriminant-validity-criteria#latest
Hello sir..i have got my, kmo value =0.512 And Sig. =0.432 so I want to know what does it means and can I go ahead with this type of values? Can I able to do cfa in my next step? please tell me And also in total variance explained , extraction sums of squared loading cumulative percentage is only 22.745 but I have seen in you told minimum 50percent is okay so what should I do now?
Those values all indicate that the set of variables you have included are not very well suited to an EFA. Perhaps you have included categorical variables by accident, or maybe you have a very low sample size, or a lot of missing data, or perhaps your constructs are formative instead.
Hi everyone. i need an answer about SEM and EFA. Is it necessary to do exploratory factor analysis to perform structural equation analysis of a theoretically determined structure?
No. EFA is not necessary unless there are newly created measures or if there is no established measurement theory (i.e., how the items should be factored).
Hi, following your video I was trying to do Exploratory factor analysis in order to get Pattern Matrix and when I did factor analysis result shows on display that 'Correlation Matrix' this matrix is not positive definite and 'total variance explained' extraction can not be done, this extraction is skipped. can you help me to solve my error please!
This happens when: 1. There is zero variance on a variable (e.g., all values are the same) 2. There is high kurtosis (e.g., mostly all the same value) 3. There are too many variables and not enough sample size 4. There are categorical variables included (e.g., marital status, religion, etc.) 5. The data is not good and so the variables do not adequately correlate Hope this helps!
it is pretty high, but if your pattern matrix is okay and your percent variance explained is okay and the commonalities are not too bad, then I would not worry too much about the 31%, especially also if you didn't constrain it to a fixed number of factors.
Most of these can be found in Hair et al 2010 (Multivariate data analysis). For anything else, here is a link to some useful references: statwiki.kolobkreations.com/index.php?title=References#Exploratory_Factor_Analysis
@@Gaskination i meant. If i have 18 items in my research based on 4 variables. 3 dependent and one independent can they overlap among items or each variable should have separate items?
@@randomentertainment5648 Ideally they should all be separate. Otherwise, it is difficult to argue that the relationships are non-tautological. It is also difficult to establish discriminant validity if the factors share indicators.
@@Gaskination thank you for your response. Its really helping me in my research. Please guide my AVE values are coming too low like.close to 0.3. Please tell me how i can increase my AVE values?
@@randomentertainment5648 AVE is calculated using the indicator regression weights. So, if those are low, then the AVE will be low. First make sure they are reflective. If not, then AVE is irrelevant. If they are reflective, then check this video: ruclips.net/video/xVl6Fg2A9GA/видео.html
Thank you for your helpful series Mr. Gaskin. Your way of expression is very comprehensive, informative and enjoyable at the same time
Thank you James for keeping this video here. I haven't done an EFA in a while and your video helped me to remember how to and why. I also had to refresh my CFA-SEM skills and again your videos helped me through the process. Many thanks
Hi James, when doing EFA, after eliminating cross loading in the pattern matrix, there are still 2 items which have loading more than 1, should I delete them as well? Thanks
These are not too much problem. EFA is just exploratory. If you want to eliminate them, use Varimax rotation.
@James Gaskin Hi James, what should I conclude if the loading is more than 1? Can I just say that the variable is strongly correlated with itself? Thanks again.
@@cunghoctienganh In an EFA, that's not a big deal. You can just leave it. If you really want to fix it, you can use Varimax rotation instead.
James Gaskin thanks, i have been watching your videos for about 5 hours today to find out the answer :D but thanks anyway. I really like your channel. It saves my life :D
Hello James, I have a very urgent question. (If anyone else can explain would be great as well!)
Why do you do EFA If you already have the theoretical constructs? In the Book of hair et al “multivariate data analysis” it states, that you only do EFA when you don’t know anything about the structure of your indicators. But if you know the structure due to theoretical backgrounds, you only do CFA to confirm these structures... I am so confused and would be so greateful if Someone could explain! Thank you!!
A non-trivial amount of debate exists among methodologists regarding whether the EFA is absolutely necessary, particularly when the set of observed measures have all been validated in prior literature (Costello and Osborne 2005), and when the factor structure is already theorized (i.e,. when we already expect, for example, that the jsat1-5 variables should factor together to represent the job satisfaction construct). As has been shown in replication studies (cf. Luttrell et al. 2017; Taylor et al. 2003), the same scales will perform differently when in the presence of other measures (not in the original study) or in the context of a different sample. Thus, in my view, the EFA should always be conducted to surface validity issues idiosyncratic to the current context. I will not resolve this debate in a footnote. My personal school of thought is that it is best to do an EFA first because discriminant validity problems manifest more visibly in the EFA than the CFA. Then follow this up with a CFA informed by the EFA. The EFA is for exploration only, and should be used mainly to highlight potential problems (such as discriminant validity) that will likely resurface in the CFA.
Hi, Professor Gaskin. Thanks for this amazing series!
In this video, you mentioned something about mitigation strategies (min. 4:30)
I couldn't find any specific reference in your list of videos, although I still have many more to go through.
Could you please indicate some specific references on mitigation strategies?
Thank you and kudos for this channel !
+Dorin Stanciu Here you go. It is queued up to the right part of the video: ruclips.net/video/XYHrmDs68Bg/видео.html&t=1h22m55s
Professor +James Gaskin , thank you! I"ve already found some tips and tricks embedded in many videos, including this one, but I felt the need for something more specific. Again, I must express my gratitude for your videos, they're some of the most useful resources on CBSEM (along with your work on PLS) that can be found on the Internet. Congrats!
Hi James, it was pointed out all reflective latent items must be included in the EFA. What about items with different level of Likert's scale? Cheers
Hi James, good job! I like the videos. Quick question - do you consider PCA to be part of EFA?
Yes. It is one of the main extraction methods.
Thank James.
Dear Pr. Gaskin, thank you for the video. I have the following question: the dataset I am running the analysis on has many 996(not sure what to answer) answers, and those are spread all over the dataset, which makes it complicated to run the analysis. Can I run multiple imputation or EM imputation to replace them with something more sound or is it falsification of the results(since technically they are not missing values) and adding more bias to the data? If not, I don't know what can safe the analysis, because if I use listwise/pairwise deletion, I end up losing more than 2/3 of the dataset...
That is too much data to impute unfortunately. I try to avoid including n/a or unsure options unless they are needed to prevent invalid responses. You can try running without those values, but if it is too little of a sample size, then you might have to run smaller EFAs (with fewer items).
Hello Dr. Gaskin.
I hope you find time to clarify the following doubt. While doing EFA as suggested I noticed that two theoretically distinct factors were loading as a single factor. When I had run an EFA on those two separately, without any item removal, they had identified as two different factors but when added to the remaining variables for an EFA, these two again load as a single factor. I understand now that I would have to do my EFA for these two separately and consider them as second order constructs and another EFA for the rest of the variables in the model. My doubt now is that should I still stick on to the threshold requirements of total variance explained (
With only two factors, the variance explained might be low. That is understandable. Same issue with the non-redundant residuals. Fewer items means each residual has more influence. So, in both cases you can be more lenient.
@@Gaskination Thank you very much for taking time to reply and clarifying my doubts 🙇♀️
Hi James
Thanks for the wonderful work
My question is
Am i supposed to delete some factors in case the EFA comes out with less factors (imagine in this case we got 5 factors instead of the 6 factors that were being expected)
Regards
Try to keep all theorized factors. You can even constrain the factor analysis to extract exactly the number you want. See if you can make that work. Also, make sure all your factors are reflective and also that they don't belong to a second order factor.
Hi James, thank you so much for these videos. They have been so incredibly helpful. Towards the end of the video you state that some literature makes a case for a Chronbach's alpha of .6 (rather than .7) especially if you have around 2-3 items. Do you by chance have a reference for this?
It is mentioned in Hair et al 2010 ("multivariate data analysis"). I don't have the book with me at home, or else I would find the page for you.
@@Gaskination Perfect thanks so much James!
Hi James, many thanks for this video. I have a complex model that consists of dependent and independent variables that are multidimensional. Can we use the same EFA method in this case? Or should we do an EFA separately for each variable?
I like to separate my higher order factors for EFA, but include them all together for CFA.
Hi Professor +James Gaksin, thanks for the fantastic videos. Really appreciate your time and effort. I have a general question concerning reliablity: is it possible to use Cronbach's Alpha as a reliability measure in a multi-group case? I expect a large amount of variation between the groups which would distort the Alpha, would it not? I would appreciate your feedback. Thanks and kind regards!
Usually multigroup effects are theorized at the structural level (relationships between constructs), rather than at the measurement level. In fact, to even test multigroup effects at the structural level, you must demonstrate no differences at the measurement level (invariance). So, cronbach's alpha should be stable even when there are multiple groups, unless you expect these multiple groups to be different at the measurement level.
Link to Formative vs. Reflective Measures in Factor Analysis: ruclips.net/video/gw0xvvJw-AM/видео.html
Dear James,
Thanks for your awesome work here. I already learns many new things!
However, I still got some Questions. For my master thesis, I want to find predictors for quality of life. To measure the quality of life (QOL) I used an already existing questionnaire with 21 Items, which are represented in 2 Dimension of QOL (eg. Mental, Physical). Do I still have to run an EFA for that questionnaire? I expected 6 Factors and got 11 when I run an EFA with all my variables. Many variables are loading in 2 factors. For example with 0,3 in 1 and 0,7 in 2. What can I do about it? Some others with -0.3 in 1 and with a value of 0,7 in another factor.
Best regardand thank you.
I always run the EFA, but sometimes you can get away without running it if you are using only established scales. With multidimensional constructs, it is sometimes best to run the EFA separately for each dimension. So, if you expected 6, that means you probably expected 3 for mental and three for physical. So, run those two group separately. Crossloadings are fine as long as they are sufficiently distant from the primary loading (I use 0.200 as my difference threshold, but others use 0.100).
Thanks you very much!! Thats helps a lot!!!
Hi James, thank you so much for these tutorials. You are walking me out hand by hand! I have a question on the EFA.
There are 12 items and the KMO is .899, Barlette 983.25, df 66, sig. the covariance of all 12 items are bigger than .3; but i loaded only 1 factor, accumulating 59.841%. Why I have found only one factor? is this suit for EFA? Thank you!
This means there is a strong correlation between all of the items. Try different factor methods (ML, PCA, PAF) and rotation methods (Promax, Direct Oblimin, Varimax), and try forcing it out to as many factors you expect. If none of this works, then consider whether method bias might be strongly influencing your data.
@@Gaskination Thank you. The instrument is from literature but I didn't see any EFA or CFA on the instrument. Can I force EFA or just do CFA? Using CFA with 12 items under one latent is acceptable? Thank you,James. I'm following your STM series step by step. Words can't express my appreciation to your help!
@@xiaoyuan1680 There is debate whether an EFA is required. The short answer is that it is not required, but it is helpful. You can constrain/force the number of factors to see if it works. If not, you can jump to a CFA to see if it is any better. Just watch out for discriminant validity issues.
Dear James. Again thanks a lot for this very useful video. In my model, I have 12 constructs(IVs, mediators and DVs). When I put them all together, only 8 factors emerge. If I exclude the very problematic scales (3 of them), I need to fix the number of factors in order to obtain 9 factors otherwise it is 7. So how should I proceed? Any reference for fixing the number of factors? Any references for any other suggestions? maybe you can upload a messy model like that and show how to deal with it.
If any of the constructs in your theory are 2nd order, then you may want to do a separate EFA for those. Otherwise, it will be difficult to argue for 12 factors (or 9). The other option is to skip the EFA for now, and go straight to CFA. Get discriminant validity sorted out during the CFA and then maybe return to EFA.
Thanks James! Very Helpful :))
I have question; What if different variables loaded in same factor ?
This happens all the time. You just need to make sure the variable loads most strongly on the factor it is supposed to load on, and that its "cross-loading" is at least 0.100 less than its primary loading (but ideally 0.200 less).
Thanks Mr. Gaskin for ur video.
btw I have question, what we have to do if we have non-redundant residual >5%?
Actually I ran EFA just like what you did in the video, but I got non-redundant residual=7%
My KMO is okay, but my "Extraction Sums of Squared Loadings" is 51%
7% is fine. Some say even up to 50% is fine. Most don't report this measure anyway. The Extraction Sums of Squared Loadings at 51% is fine. It is low, but not too low. Anything less than 50% is discouraged.
Thanks a lot Mr. Gaskin
Hello professor. While we have 3 variables, and each variable consist 3-5 items. It is necessary to do EFA with all items, so it is going to be reflective on its variable? Or it is a must to do EFA one by one for each variable?
It is best to include all items of all reflective constructs in the EFA. This will then allow you to assess discriminant validty. If tested individually, you cannot assess discriminant validity.
I have a query related to the reliability of items. I have a latent variable that consists on two observable items. These items are taken from past literature. One item is reverse coded in previous papers. When I tried to reverse code it, it is showing the reliability of 0.58 and I tried both ways of coding but the reliability still remains same of 0.58 in both cases. Why?
reverse-coded items are notoriously bad for reliability. Two items, with one reversed, is a recipe for poor measurement.
Hi!! thanks a lot for the video!! A quick question plz: if I have a formative construct with first-order reflective dimensions, can I use a PCA or an EFA in SPSS?
You can use either approach to validate the reflective first-order dimensions.
@@Gaskination Thank you very much!!
Hi James, how about goodness of fit test which indicates the difference between the estimated and sample correlations, and thus indicates the factor model is a good representation of the relationships amongst the observed variables or not?
During the EFA, you can assess this if you use Maximum Likelihood approach. it will produce a goodness of fit measure and the degrees of freedom that you can use as a denominator. If the quotient is between 2 and 5, then you're in pretty good shape.
Hello James, I have no cross-loadings in the pattern matrix; however, 2 additional factors have been created mainly due to one variable, which is making 3 different factors. What can be the issue and solution for this?
Sounds like it is a higher order factor with three dimensions.
Hi James! Thanks for the informative video. I noticed that in my EFA, one of my variables loaded above 1, is this a problem? I found that when I changed the rotation to Oblimin, it reduced from 1 to .718. Is that a problem? Furthermore, I did have separate loadings without any cross loadings, but unfortunately a variable I theorized (identity affirmation) loading neatly onto my mental health latent variable instead of what I originally theorized (under resilience strategies). As such, is it fair to drop it entirely from my analysis?
The heywood case isn't much of a problem during the EFA, since it is only exploratory. As you note, you can usually hide it by changing the rotation method. When a set of indicators load with another set (unexpectedly), I would recommend running a separate EFA with just those items (from each construct), to see if you can separate them by removing one item at a time. If they cannot be separated, then consider whether you might have a 2nd order factor with multiple dimensions.
@@Gaskination Thank you so much, I definitely will do that :) Thanks!
Hi James,
I have a question to ask. My conceptual framework is unlike yours where 1 observed variable has 1 item. Each of my observed variables have at least 1 items. I do not know how to use EFA in this case.
For example my latent variable consists of 3 observed variable each with 2 items.
If a variable does not have more than one indicator/item, then it is not latent, and therefore does not belong in the EFA or CFA. You can bring these single (observed) items in during the causal model.
i have one question you did EFA for all scales together. I have 5 scales in my research can i do EFA for every scale separately?
If you do it separately, then you cannot assess discriminant validity. Discriminant validity assessment is the primary strength of the EFA. So, I would strongly recommend a combined EFA.
Dear James... I am building a questionnaire on videogames as a part of my thesis. I have 8 variables under 4 blocks, each with 4 items: intrinsic motivation towards games (based on a scale by another author), then three varaibles that have to do with player needs (autonomy, competence and relateness) based on a scale by another author and finally a block on importance given to game elements (conventions, components, actions and emotions). Im not doing SEM and EFA and CFA are only to validate the pilot.
Should I do factor and confirmatory for each block independently as they are different? Even though It will be easier (avoiding cross loadings, etc.), I think it would be more representative if I do them together at least the first two blocks together so I can be sure I am measuring what I want, isnt it? Big thanks!
It is best to include all reflective latent factors in the same factor analysis so that you can be sure the measures are distinct (discriminant).
Thanks so much! Still, to my understanding, if there are particular scales within the questionnaire (like the one that have to do with player needs), in the confirmatory you want to check that this scale is "three-factorial" and you should run it independently, don't you think?
No. I would still run the CFA with everything else, just as a second order factor. However, in the EFA, I would run it separately.
Hi James, Thank you for your efforts. I have a problem. I am running EFA for a framework that consists of multidimensional constructs. i.e. i have technology factors that include sub-variables such as security, privacy, trust and other multidimensional such as organizational factors which include top management support and organizational readiness. My question: is it right to run EFA separately for the organizational factors and then for the technology factors? if yes, is there any reference to support that? Thank you very much!
It is right to run them separately, but I'm not sure if there is a good reference for this. If you find one, please let me know and I'll put it on the statwiki.
Thank you. I only found indication in Hair et al. (2010, p.101) referring to the fact that it is inappropriate to run the IV and the DV in one EFA. I found also other literature but not really reliable. I think it is a good idea to write paper about it and to compare full EFA with separated EFA.
Thanks a lot!
Hi James,
I am using a 33-item measure (Driving Anger Scale) and using promax rotation leaves me with quite a few loadings below .4. Would it be fine to use varimax? It has provided me with stronger factor loadings.
Cheers
Yes. That's fine. According to some, the rotation method doesn't matter very much.
what if the reproduced correlations show >5% nonredundant residuals?
Hi Mr Gaskin. Good Morning
How to perform EFA if DV dont have any sub-dimensions? Can we run EFA for DV along with any other construct?
Question.2. Is there any need to perform EFA if construct has only 3 items (with no sub-dimension)?
Definitely collectively. This is so that you can determine the discriminant validity between variables. This is the primary purpose of an EFA (to show factors and how they are distinct). So, doing an EFA with only a single factor undermines the purpose of the EFA. No reference needed since this should be common practice. You can also examine any article using SEM in a top journal. They should follow this approach.
Hello professor Gaskin
Thank you for this informative video. Sir one question. Do we need to run EFA on established scale or can i directly go for confirmatory analysis directly??? Sir if you have any reference that justify applying CFA directly without EFA. Please share it with me 🙏 I'll be forever great full to you as I have run CFA without running EFA in my dissertation.
EFA is not necessary if all measures are already validated in prior literature. This is established process, so no citation needed. If you really want to cite someone for it, you could cite Hair et al 2010 (Multivariate data analysis), as there is a whole chapter on EFA.
@@Gaskination thank you so much for the reply sir. 🙏
@@Gaskination my lecturer said "The prior theory validity will be different if you are doing research in different places, then you should go for EFA first" is it true Mr. Gaskin?
@@brantazconflix3873 I agree. Even with validated scales, an EFA can be useful if studying a new context/population. However, strictly speaking, the EFA is primarily used for new scale validation, rather than re-validation of established scales. Most would probably suggest to go straight to the CFA for established scales.
Hi, James, is there a reference you could provide me in regards to your determination for an acceptable amount of non-redundant residuals with absolute values greater than 0.05, and why there is a recommended 5% ceiling? - Thank you!
hmm... I've never really thought about that particular threshold. It is just what I was taught so long ago. My guess is that it comes from Hair et al 2010, but if not, then maybe from one of the ones on this list: statwiki.kolobkreations.com/index.php?title=References#Exploratory_Factor_Analysis
Thank you. I'll check them out!
Dear Prof Gaskin, I have a question: items from two different constructs are loading on the same factor, what should it means and what should I do
I have found the best way to resolve this type of issue is to do a separate EFA with just the items from the offending factors. Work out this smaller EFA (by removing items one at a time that have the worst cross-loadings), then reinsert the remaining items into the full EFA. This will usually resolve the issue. If it doesn't, then consider whether these two factors are actually just two dimensions or manifestations of some higher order factor. If this is the case, then you might consider doing the EFA for this higher order factor separate from all the items belonging to first order factors. Then during the CFA, make sure to model the higher order factor properly by making a 2nd order latent variable.
Hello Dr. Gaskin, I have a question regarding running an EFA. When I run mine, it says that "Rotation failed to converge in 25 iterations." My KMO and Bartlett's test as well as communalities are acceptable, so I'm wondering what could be happening?
It is most likely either low sample size (compared to number of variables), or there are one or more variables that don't belong (e.g., formative indicators, categorical indicators, grouping variables).
I experienced this. For the same data set, I allowed SPSS to determine the number of factors and came out with 9. I tried to set the factors at 5, I got the same message- rotation failed to converge in 25 iterations. I tried 6 factors and it worked. I surmised that maybe at 5 factors, the %cumulative variance is less than 50% under the rotated variance explained column.
Hello sir, there's a question for you. What does a negative factor loading mean? Should it be reported in the matrix while reporting the results?
it means an inverse relationship with the other items on the factor. Usually this means the item is reverse-coded and needs to be re-reversed.
Hi James! Can I give my questionnaire to my experimental and control groups even BEFORE conducting EFA?
Due to lack of time to learn and run EFA, the questionnaire I gave to my experimental and control groups is not yet reduced and NO identified factors are determined as of yet.
My plan is to conduct EFA later on. When reduced items are already available, that's the time that I will be selecting appropriate responses from my study groups for my analysis. Please comment. Thank you
Yes, that is totally fine. Most people do the EFA on the questionnaire data.
Thanks for your reply. I asked the question because, initially, I thought EFA should be done first before giving it to my study groups (experimental and control).
Anyway, I will be doing EFA later. I will be analyzing (t-test, etc) my study groups' responses BUT I will be removing their responses to some items in the questionnaire that will be excluded by the EFA process.
All factor loadings of one construct are negative in my Pattern Matrix of EFA, does this happen due to the reason that this variable is negatively related to the dependent variable? Factor loadings of all other constructs are positive in the pattern matrix. All other results are fulfilling the criteria
Here is a post I've written about this: gaskination.com/forum/discussion/144/negative-loadings-in-pattern-matrix#latest
@@Gaskination Thank you very much Sir
Hiii james whould should be the maximum and minimum value of correct item total correlation
Correlations should be between +/- 1.00
Thank you for an informative video. I have one formative model, what do I need to do? if ı can not do EFA
For formative models, use SmartPLS instead. I have lots of SmartPLS videos for reference.
What are the causes of low factor loading of items even they have good reliability between the items during EFA?
Reliability is estimated in a silo (ignoring all observed measures not in that particular set). EFA accounts for variance shared across all measures included in the EFA. This means that if those particular measures share a lot of variance with other measures, then the amount of variance loading on their home factor may decrease. This is a lot like the difference between correlation coefficients and regression coefficients.
@@Gaskination thanks for the clarification
Hi Mr Gaskin, you mentioned about scores, could you please explain a little more about it. How can we use it? if you have a video, kindly share the link.
Factor scores are a weighted average of the items that have loaded onto the factor. These scores are weighted by the indicator weight (or loading) shown in the pattern matrix of the EFA. I don't have a video for it, but the procedure is pretty easy. Once you have a good EFA solution, redo the EFA exactly as it is, but go to the scores button in the EFA menu, and check the box for save scores as variables. When you run the EFA this time, it will create these factor scores as new variables in your dataset. Each factor score will represent the items that loaded on it as their primary factor.
Would you kindly provide a link where you talk about mitigation strategies if the number of factor extracted do not much your model ? (Part 4:30 of the video)
Sorry for the delayed reply. I've been busy trying to fix the StatWiki. It is fixed now though. As for your question, you can check out the 'messy EFA' video: ruclips.net/video/oeoTpXiSncc/видео.html
I have 5 constructs in my model. In used 4 "adopted" measurement scales and 1 "adapted" measurement scale for analysis. Do I need to perform EFA for all or only for the one which I "adapted". any reference article will also help along with your kind suggestions. thanks a lot
It is always best to do the EFA on all reflective latent scales, even if they have been validated previously in the literature. The EFA allows you to detect discriminant validity issues specific to your set of constructs, and convergent validity issues specific to your sample. As for a reference, here are several: statwiki.kolobkreations.com/index.php?title=References#Exploratory_Factor_Analysis
Got your point. Thanks a lot for your feedback. So kind of you.
Hello Sir, Can you elaborate about the 'Measure' lile Scale, Ordinal and Nominal. I am working on Theory of Planned Behavior. What do you suggest the 'Measure' to these Psychological Variables, i.e. Scale or Ordinal.....
Thanks..Waiting for reply
Usually TPB constructs are measured on ordinal scales. Ordinal is discrete interval scales (such as Likert e.g., Strongly Disagree to Strongly Agree). Continuous is not discrete (such as age, income, years of education, etc.).
@@Gaskination Thanks for the Information..
Hai sir, can you please tell me the time stamp in this video when you do "Convergent validity"? Thankyou☺
Sir,
I carried out EFA and got 6 factors as per the model.
Kindly guide me on how to interpret the results in APA style. I am searching for complete interpretation of results obtained from EFA, CFA and SEM.
Please help me.
I know nothing about APA. Sorry about that.
Hi Prof James,
Is there any video example of doing EFA for higher order model e.g.: 2nd order model?
I've tried to do the EFA based on this video but the pattern matrix always show 'rotation failed to converge in 25 iterantions'
Please let me know what did I do wrong. Thanks
For models with higher order factors, I usually conduct a separate EFA for the items of the higher order factor (so that just its subdimensions will factor) or I skip EFA (if the measures have been validated already in extent literature), and then just go straight to the CFA. Another approach is to use principle axis factoring instead of PCA or ML.
@@Gaskination Hi Mr. Gaskin,
Thank you for your reply.
The reason I need to run EFA is to solve a discriminant validity issue of one of the variables as AVE
@@arifnugroho73 You can be less strict on discriminant validity between dimensions of a higher order factor. These dimensions are expected to be highly correlated since they are manifestations of the same thing.
Dr. Gaskin - can you share the outline you are referencing? TIA
It's just on the StatWiki here: statwiki.kolobkreations.com/index.php?title=Guidelines#Order_of_Operations_for_Testing_your_Model
@@Gaskination Outstanding! Thank you
Hi James,
Can you conduct exploratory factor analysis (EFA) followed by a confirmatory factor analysis (CFA) on the same dataset?
Thanks!
Yes, when sample size is not sufficient to use separate datasets. It is considered more rigorous to use separate, but very few studies actually have sufficiently large samples to accommodate separate analyses.
Thanks, James!
@@Gaskination can i have literature on it sir??
@@brantazconflix3873 The "Multivariate Data Analysis" book by Hair et al is a good source for factor analysis.
I have 7 latent variables in the model. I took all the observable items from the literature. Still, there are 2 latent variables that are loading in the same column pattern matrix. 6 factors are being extracted. They should load on different factor columns. What can be the solution? Is it sample size or there any other thing that is causing this issue? Waiting for your reply..
You can try to constrain it to seven factors by extracting based on a fixed number rather than using eigenvalues. (check 2:22)
@@Gaskination Why the Pattern Matrix wouldn't able to differentiate the 2 factors in column matrix? Is there a chance that the observed variables are less for that certain latent variable to differentiate among them? Can you elaborate a little bit about it as the variables are already taken from the literature? Thanks in advance.
@@abdulmoeed4661 Here is a post I wrote about it: gaskination.com/forum/discussion/98/what-to-do-if-two-factors-dont-pass-discriminant-validity-criteria#latest
Hello sir..i have got my,
kmo value =0.512
And Sig. =0.432 so I want to know what does it means and can I go ahead with this type of values? Can I able to do cfa in my next step? please tell me
And also in total variance explained , extraction sums of squared loading cumulative percentage is only 22.745 but I have seen in you told minimum 50percent is okay so what should I do now?
Those values all indicate that the set of variables you have included are not very well suited to an EFA. Perhaps you have included categorical variables by accident, or maybe you have a very low sample size, or a lot of missing data, or perhaps your constructs are formative instead.
Hi everyone. i need an answer about SEM and EFA. Is it necessary to do exploratory factor analysis to perform structural equation analysis of a theoretically determined structure?
No. EFA is not necessary unless there are newly created measures or if there is no established measurement theory (i.e., how the items should be factored).
Hi, following your video I was trying to do Exploratory factor analysis in order to get Pattern Matrix and when I did factor analysis result shows on display that 'Correlation Matrix' this matrix is not positive definite and 'total variance explained' extraction can not be done, this extraction is skipped. can you help me to solve my error please!
This happens when:
1. There is zero variance on a variable (e.g., all values are the same)
2. There is high kurtosis (e.g., mostly all the same value)
3. There are too many variables and not enough sample size
4. There are categorical variables included (e.g., marital status, religion, etc.)
5. The data is not good and so the variables do not adequately correlate
Hope this helps!
Hello James are you able to send a link for that data you are using ,, want to do practice am a student preparing for my thesis..
The data is on the homepage of statwiki.gaskination.com/
@@Gaskination thankyou professor,, you have made learning very interesting through your links
Dear Sir,
when i did EFA , I got a non redundant residuals of 31 %. I there any problem in that? And if there is any problem, how to reduce it?
it is pretty high, but if your pattern matrix is okay and your percent variance explained is okay and the commonalities are not too bad, then I would not worry too much about the 31%, especially also if you didn't constrain it to a fixed number of factors.
thank you very much sir
Dear Sir, Can you please give references for EFA cutoffs like KMO, Cronbach's alpha etc
Most of these can be found in Hair et al 2010 (Multivariate data analysis). For anything else, here is a link to some useful references: statwiki.kolobkreations.com/index.php?title=References#Exploratory_Factor_Analysis
can we get the data source just reuse it for practicing
It is on the homepage of statwiki statwiki.gaskination.com/
Where can I get this same data?
The data is available on the homepage of statwiki: statwiki.gaskination.com/
Hi james can my variables overlap with the items?
Do you mean that you want multiple latent factors to be attached to the same items? Yes, this is possible in a MTMM model or in a bifactor model.
@@Gaskination i meant. If i have 18 items in my research based on 4 variables. 3 dependent and one independent can they overlap among items or each variable should have separate items?
@@randomentertainment5648 Ideally they should all be separate. Otherwise, it is difficult to argue that the relationships are non-tautological. It is also difficult to establish discriminant validity if the factors share indicators.
@@Gaskination thank you for your response. Its really helping me in my research. Please guide my AVE values are coming too low like.close to 0.3. Please tell me how i can increase my AVE values?
@@randomentertainment5648 AVE is calculated using the indicator regression weights. So, if those are low, then the AVE will be low. First make sure they are reflective. If not, then AVE is irrelevant. If they are reflective, then check this video: ruclips.net/video/xVl6Fg2A9GA/видео.html