Here's a fun pet project I've been working on: udreamed.com/. It is a dream analytics app. Here is the RUclips channel where we post a new video almost three times per week: ruclips.net/channel/UCiujxblFduQz8V4xHjMzyzQ 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!
Sir James Gaskin. Sir, how to create syntax plugin as like pattern matrix builder in AMOS and then if there have problem in error amos ; Assertion Failed Finalizer CDbfile.id=dbGraphics di CDbfile.Finalize(). How to handle it, please help me, Thanks...
@@dedylesmana The way to create a plugin is by writing the code in the AMOS plugin editor. It can be written in C# or VB.net I think. As for troubleshooting that error, it could occur for many reasons, some of which are explained here: statwiki.gaskination.com/index.php?title=Plugins#Troubleshooting
@@dedylesmana I don't think I understand your question. Here is a troubleshooting guide that might help: statwiki.gaskination.com/index.php?title=Plugins#Troubleshooting
CFA serves many of the same purposes, but not all. EFA is good for unconstrained factor analysis, in which you are not assigning items to factors. CFA assigns items to factors, and allows for more robust tests of discriminant validity. You can also test common method bias and group invariance better during the CFA.
No, I do not have a lot missing data and not using reverse-coded items. I would check with your video soon and see what happen next. Btw, thank you very much James for your help. Really appreciate much your help with all these videos. Excellent work!
2) After calculating the model and showing the numbers, you say the value should be above .7 but I have a few below. However I think the ones lower are where factors where 2 factors in EFA loaded together and I wanted CFA with the same model before trying my pre EFA model or even deleting some items to solve the cross loading. Thanks a lot for your time!
Estimation issues. Might have negative error variances. Check standard errors for anomalies. Watch my video on "iteration limit reached in amos" to get some ideas.
Hello James entire SEM series is really helpful for me . I am badly stuck up with one problem for a particular model where its showing Iteration limit reached . I kept Iteration limit as 50 . But the results for fit indices are satisfactory .
Look at the Notes for the Model section of the output to see if it gives any clues. You can also watch my "iteration limit reached in amos" video for some additional clues.
Hi Sanjukta, RUclips isn't letting me reply to your comment directly, so I will reply here. I have a video that discusses "iteration limit reached in amos" Watch that for some ideas. I also go over it in my bootcamp videos when I talk about common method bias.
I have a question regarding the Experience high and low and frequency, how you did calculate them in spss at first and also how i can identify mine. Btw, thanks a lot for your great and helpful videos.
James Gaskin OK, my moderators are for example technology level, and frequency usage of that technology, I have those in my SPSS data as multiple answer questions since they are Likert scale 1 to 7 style. sp how i can group them first to 0 (1-3) low, then 1 (4-7) high, then i can chose them in my Amos. Also in the part where you have CLP the common bias, u had a negative results from joy but in my case i had them with some of my variables with CLP, so how can i fix that, because i tried your moving the parameter way, but still, do i delete those negative loading factors ??? or what? I'm so grateful for your generous help. Thanks a lot.
Hi I installed the plugin but when I paste the pattern matrix it is really messy therefore the diagram is not correct as well. I would appreciate if you help me to solve this problem . thank you
I would recommend resolving the EFA first, then doing a CFA. If the matrix is messy from crossloadings, then just suppress loadings less than 0.300. The plugin also is known to fail if you are using comma notation instead of decimal notation (e.g., 0,05 instead of 0.05) or if you are using variable labels instead of variable names in the EFA.
No. Since word of mouth is not latent, it does not belong in the CFA. The CFA is only for latent variables. You can add word of mouth when you start testing a causal model.
Hi James, the Excel file is very helpful, however, when I am writing the thesis, how do I make justification of using this file as I think the examiner will not accept it if I say I downloaded from the internet. Thanks James.
You can cite it or use one of the updated plugins and cite that. How to cite the excel file and the plugins is shown on the homepage of statwiki. Thanks!
I've never seen an error with that specific wording, but it sounds like a variable with that exact name is not appearing in the specific dataset you linked in AMOS. You might have linked the wrong dataset or spelled the variable name incorrectly.
I would not drop the factors so readily. I would first fight for their survival. See if there are data normality issues, outliers, convergent validity issues, etc. Make sure they work in the EFA after that. Then, after all of that, test for reliability. If they are close to 0.700, move forward. Drop as little as possible.
HI James! So, VERY NEW to these procedures. I don't understand where the pattern matrix comes from if SPSS only computes EFA or PCA. Thanks for providing feedback on this question that is probably silly.
Hi James, thanks a lot. I have a question please, What if the results of EFA were different from those of theory? we conduct CFA basing on which of them? Let's assume that we have a concept "attitude" which theoretically has three dimensions (cognitive, affective & behavioral), but the results of EFA shows only one factor. In this case should we ignore EFA & treat Attitude as a second order factor withe three dimensions? Or we should respect the EFA results & treat it as a single one order factor? I have such problem in my research & I couldn't figure out what to do.
Hi James. Need to ask.. Say that I have 5 factors in EFA and turned out that only the first 2 factors are reliable and the other 3 factors need to be deleted (Cronbach's alpha
Thanks for your fantastic videos, not everybody is willing to share these knowledge for free. Like what you always say, SMILE :-) , That's what I do when I watch your videos. One question, is there any way to use the stat tool package for a variable with three group, like low, medium and high?
James, can you talk about the rationale behind the use of modification indices in a confirmatory framework. Specifically, why look at the residual covariances? Or, the larger question, are we actually doing theory driven work if we are relying on data output to improve model fit? Thoughts?
Model fit is all about accounting for all the major covariances among the set of variables in your model. If they are accounted for and you still have strong loadings and good discriminant validity, then we are satisfied that we can rely on the measurement model. Thus, addressing modification indices and standardized residual covariances is not warping our model or fishing around. It is simply "controlling" for major covariances we didn't theorize, in order to validate the covariances we did theorize. I hope that helps.
great work James! when I did EFA I received this error Warnings Text: Q63_Comm An undefined variable name, or a scratch or system variable was specified in a variable list which accepts only standard variables. Check spelling and verify the existence of this variable. This command not executed. I have 10 variables with 85 items. would you please help me and explain how can I solve this error?
Hello Mr. Gaskin, I have been trying to use the stats tool package suggested by you, however after loading the data onto the downloaded excel sheet, it does't work and displays a message saying 'debug'. How can I avail the correct version of the software? Thank You.
Shivangi Sanganeria Make sure to look at the caveats and assumptions. If it still doesn't work, then email it to me and I'll take a look. james.gaskin@byu.edu
But in the Stats Tool Package there are only two boxes for groups to paste the tables. I think we have to add something in the Visual Basic which I am not expert in it.
Did all that, checked for the normality issues, outliers, convergent validity issues and all that.. They surely did work in EFA but still are not reliable (Not even close to .7, ranges from 0.5 - 0.6)
Thanks a lot for yet another great video. I am walkung step by step with you and everything is looking fine. I have two quick questions if I may: 1) The modification indices area only have some factors under the regression part. Do I need to remedy these? All of them dissappeared after I created groups for invariances for age and experience even when options still show them as on.
When analysing the common method bias; if you have negative variables/values, does this mean the same as if they are low i.e. that we have differences in standard regression weights when adding the common latent variable or does negative variables/values mean that these are not acceptable at all? I have tried to move the regression weight to other indicators however, this doesnt seem to work on the results either.
Hi James, Many thanks for such wonderful video series. May I ask about a problem I am encountering. When I run CFA analysis, an error message popped up saying that "Either the constraints on the parameters or the initial parameter values are bad. This error can occur, for example, if you request a maximum likelihood solution while specifying initial parameter values that prevent the population covariance matrix from being positive definite. Check to make sure that every endogenous variable has a residual variable (error variable) associated with it". I am using your data file, go step by step following your videos. Could you please help advise how to fix the problem? Great thanks James Huyen
Hello James, I have a problem performing CFA. Every time when i run the analysis, Standardized Regression Weights are not calculated. What can be the problem?
Hi James, i solved the problem,I constrained the error variance to a fixed value. previously, it was fixed at zero and now i made it to be 1. Now, it shows the standardized values. Thanks anyway. Your videos are very useful. all the best
Hi James, Thanks for the video, which is quite helpful. ONE question: what do you think of the interpretation of residual correlation in CFA? Some scholars said in managment practice, error term corrleation is not recommended or welcomed by good journals; some others said it may indicate content overlapping in items. I find it hard to explain when validating a scale. Thanks.
Thank you, professor Gaskin! I wasn't aware of that video! It is very helpful! I have just one question. Namely I have noticed that in the video, in your Notes for Model part of the output, the probability level is 0.433, while in your SEM videos it always is 0.000. I know it is related to the sample size, but is there a rule of thumb to know when should I accept significant or nonsignificant chi-square?
Hi James. I have a question, please. I'd like know the source (article, book, or chapter) where I can find the recommendation about the use of the comparison of critical ratios criteria in the group comparison/differences (11:35 to 11:43). I'll use this criterion in my analysis. Thanks.
It is not very well supported in the literature. Here is an updated video with more support (via chi-square difference testing): ruclips.net/video/Pakg3PlppuY/видео.html
Hi James!! I am trying to run my first SEM with AMOS for my thesis and I am having some trouble to proceed with it...The first problem is that I have a model with 13 latent variables and 52 observed variables, and I do not know if I have to make the first step that you make in this video with all of them and covariating all 13 constructs (as the graphic seems kind of crazy and it doesnt let me run the analys...coul you give me a hand with that? On the other hand, I get these kinds of errors when I try to run the complete model (including unidirection arrows from the independent to the dependent variabes of my model: - The covariance matrix is not positive definite: whta does this mean and how can I solve it? - The model is probably unidentified. In order to achieve identifiability, it will probably be necessary to impose 1 additional constraint ( I have fixed 1 in all arrows from errors to observed variables and in 1 of the arrows that go from each latent variable to its observed variables...what more constraints can I make? - an error occured while attemting to fit the model either the constraints on the initial parameters or the iniciatl parameter values are bad (...). What does it mena and what can I do to be able to run the analysis withouth this error message? I AM QUITE DESPERATE AS I AM RUNNIGN OUT OF TIME WITH MY THESIS, AND SOI I WOULD REALLY APPRECIATE YOUR HELP I love you videos by the way!! They are super useful and clear for beginners lke me ;) Elisa
Elisa López Sorry for the delay. I was hosting an SEM Boot Camp last week and then yesterday was my wife’s birthday. Today is my first day back… To answer your questions: 1. Yes, all the latent factors should be covaried. The reason it won't run might be due to sample size. You have a fairly complex model. If the sample size is not greater than about 200, you will have trouble running it. If you do have enough sample size, then it could just be something about the way it is drawn. Try recreating it with my EFA-CFA plugin available on my wiki. Here is a video about how to get it: ruclips.net/video/sLtMOFcojZY/видео.html 2. The errors you have are also likely due to an error in drawing the model. Again, I recommend the plugin as it removes the possibility of creating such errors. If you still cannot get it to work, then you can email me your .sav and .amw files and I can look at it.
Hello, I want to ask about the modification indices. I have two observed variables that seem to have high scores, sort of like in the video. You mentioned that covaried them, but I don't understand the reason for doing so. What does this do and how can be it explained?
It accounts for the correlation between their errors. Model fit is all about accounting for major covariances (or correlations). So, if there is a major correlation between two variables, or errors, then you can account for that with a covariance arrow. When we do this with error terms, this means we want to account for their correlation without explicitly theorizing about their relationship.
Dear James, I proposed the following model (regarding service recovery): (interactional justice, procedural justice & disutributive justice) lead to (transaction specific satisfaction & overal firm satisfaction) leading to (repeat purchasing & word of mouth). All 7 variables are tested with 3 or 4 items (in a survey). However, word of mouth is tested using only 1 item. When performing the CFA (step 5.1), should I also include word of mouth in the model, and if so, how can I do that?
Hello, I currently have SPSS Amos version 26, I made the indications of the file "FIX Syntax Error with Pattern matrix 24 plugin.doc" to copy the file "PatternMatrixBuilderAMOSv24.dll", but I get an error, what should I do?
Dear Prof Gaskin, I run a CFA but all the regressions are unidentified. The model has 2 latent variables with 5 indicators for one of them and 6 for the other. I run an EFA before the CFA and get rid of cross loadings. The reliability is also in the accepted rates. The bivariate correlation between the two latent variables is 0.578. The sample size is 216 cases. Could you help finding the problem? Thanks
bayan khalifa Check the notes for the model in the output to see if there are any clues there. For example, it might say you are missing a constraint, or that you reached the iteration limit.
Hi James and thank you for this tutorial on SEM. It is extremely useful for me. I'm developing a model and I got these results for the default model in AMOS: Result (Default model) Minimum was achieved Chi-square = 4,252 Degrees of freedom = 4 Probability level = ,373 How do I report this? Is this a good model to report or does it have to have a probability level < ,05? I'm really confused with this probability level. Other results for this default model are CMIN/DF 1,063, CFI ,999, PCLOSE ,881 and RMSEA ,010. If I understand correctly, all these results show a good model? Thank you so much!
Hi James! Thanks for your videos, they have been very useful for me. I have a question, my Amos Output don't have "Corelations" and "Covariances" under Estimates - Scalars. Do you know what can be the problem?
Hi Dr Gaskin, After completing the process of screening of data and EFA, when run the CFA initial model, I got this error 'GROUP 1, check for incomplete data', I followed your each step and used the same data as you did, but don't know what happened? please help.
Hi James, am following your videos and I am just wondering why one would do a CFA after an EFA on the same dataset? I have seen papers published with just the EFA done to show factor structure. I thought the main reason to do the CFA would be for new samples?
Hello James, I have been following your videos and have found them amazingly helpful! But I ran into a little snag when trying to paste my Pattern Matrix from SPSS into the AMOS Pattern Matrix Input plugin window - I can select and copy the Pattern Matrix, but I can't paste it into the pattern matrix builder window in AMOS (when I right mouse click the 'paste' option is grey/unavailable & using paste short-cut key does not work either). I believe that I have correctly installed the plugin in AMOS (the window opens and all looks okay) & there is no problem with copying the pattern matrix (i.e., I can paste it into a word doc), but it's just not allowing me to paste into the Pattern Matrix Input window. Have you come across this problem before, and/or do you have any possible solutions that I might try to fix it? Many thanks, Kim
Kim Woodward I haven't come across this before. The only reason AMOS would not let you paste would be because it does not recognize that anything is on the clipboard. Make sure to copy the pattern matrix before opening the plugin. If this is what you did, then try copying after opening the plugin. If this still doesn't work, then I'm not sure. I suppose you'll have to do it the good old fashioned way... Here is a video where I show how to build it from scratch: ruclips.net/video/JkZGWUUjdLg/видео.html
Hi James - you're saving so many PhDs, I am sure :) For some reason, in my output for the invariance test, the 'Pairwise Parameter Comparisons' isn't there? I've gone through the video a couple of times to see if I've missed checking something, but cannot see that I have. Anything else I can check? TIA (thanks in advance)
+Jennifer Barhorst It should show up if you have multiple groups assigned, if you have assigned those groups data, and if you have checked the box for critical ratios for differences.
Hi James, Thank you so much for this video. I run my model and I get the results. The results showed that one out of 5 factors in my questionnaire correlated negatively with another factor. Is that mean the questionnaire not valid? Is there such way to sort out this issue? Thanks
Sorry to bother you again. At first, my model showed a bad fit. The fit became acceptable after some modification. However, after adding the CLF the model fit has increased again. If I have to report model fit in a paper, at what point in time do I measure fit?
Dear James İ have done a CFA and eliminate some factors and draw some correlations according to your youtube lessons and achive acceptable goodness of fit but validity and relibility is very low. what should İ do.? should İ only say that the model is unvalidated or İ have other ways? REGARDS
Dear Prof. James Gaskin. Is it ok if I use your excel tool to compare MM between groups with un-arranged indicators? just copy it to excel, and it can proceed? Thanks Prof
I don't think I understand the question. However, my excel tools are designed for AMOS output only. They are very NOT robust. The tools expect the exact structure of the AMOS output.
Hi, Is there anywhere in this series or your other videos where you discuss how to use observed variables of different measurement types as indicators of the same latent variable?
Not that I can think of. If you mean they are of different scales, then that is fine. If you mean they are of different dimensions of the factor, then you will want to use formative modeling.
Hi James!, Can you tell me is your plugin works on new version of Amos (v24), because I have got a problem with loading it. I will be grateful for your answer :)
Hello, professor Gaskin. Thank you for your videos and an extremely helpful explanations and examples. I have only one issue when performing the common method bias. Namely, I get the error "A sample of parameter values was inadmissible" and my new data set with composites is not imputed. Do you have an advice on that? Many thanks.
Hi James, I'm following your videos step by step. My final table in the GroupDifferenceTab has all the columns equals to the ones shown in the video apart from the z-score column. I don't understand why all the variables have a sig. value (0.000). What could have happened?
Hi James my name is Emile I am a PhD students in South Africa. I find your video very interesting. I have been asked to analyses the "Factor structure of PsyCap and OCB, internal consistency of each of the sub scales, equivalent unbais measuring across different ethnic groups".Psycap and OCB are two constructs with sub construct; I was wondering if this is the right video to watch could you give me some advises on how to proceed? Thank you
Thank you. This was very helpful. I only have one question in regards to the video. As you showed; you need the pattern matrix however, when I did my factor analysis in SPSS by making use of extraction method: principal component analysis and rotation method: varimax, I did not recieve a pattern matrix. Does this not allow me to do an CFA or can I use something other than the pattern matrix?
Hello,I was following your example trying to use one of my previous pattern matrix, however, I cannot get the file to upload in AMOS. What am I doing wrong?
Hi James, I am comparing samples from two different countries. Please guide me in the following isues: Do i need to perform CFA on different data sets separately or i can combine them? While running the structured model how do I compare two different populations? Is it Fine, If factor loading are more than 1 among variables and constructs in CFA results ?
Dear James İ want to apply a distincted model (İ Have taken it from another research) and research the validaty and reliability of that on a new sample . but İ dont know what İ should do? İ think İ should do efa before cfa but the problem is that the the results of efa is very defferent with majour model, please reply me if İ should do efa at firs or cfa ?
I always do EFA first if I have new data, even if the constructs are all well-established in the literature. The EFA reveals discriminant validity issues much better than the CFA can.
Hello James, I used to Pattern Matrix Builder to paste my EFA but the loadings were running, the display didn't come out as yours did. How I can make it look like yours?
Daisy G I'm not sure what you mean by "running". Sometimes the plugin doesn't work when you use variable labels instead of variable names in SPSS. It also has been known to fail when using commas instead of decimals (as done in many European countries). If all else fails, you can try to build it by hand as shown in this video: ruclips.net/video/JkZGWUUjdLg/видео.html
James Gaskin Thank you so much. I still can't figure out why. As I cut and paste the EFA into pattern matrix builder, the arrow of the error values are not in the right place. And the whole model doesn't space like yours in the video. I am gonna try changing that variable labels and see if that works. Thank you.
Hi Prof.James, when I tried to run my CFA model. I kept getting this error message that "the variable, "nameofvariable" is represented by a rectangle in the path diagram, but it is not an observed variable". I hope that you would be able to enlight me on What could have been the cause of this error? I am not sure how to deal with this as I am quite sure that the variable is an observed variable. thanks.
+Law jiawern This can happen if it is spelled or spaced incorrectly, or if you accidentally uploaded an earlier version of the dataset or the wrong dataset. A way to prevent this is to simply go to view, variables in the dataset, and then drag the variable into the rectangle.
Dear James, I am following this video with a similar model, however I have some problems. When I run my CFA, most values between the factors and items are higher than 1. Is that possible? Aren´t they supposed to be between -1 and 1? Futher I above my facotors there are values between .99 and 2.35, which I cannot spot in your model. Is that an additional setting to be taken before I run the model? Thank you very much for your help and the great video series in general! :-)
The factor loadings should be between -1/+1, but my guess is that you are looking at UNstandardized loadings. You need to check the box for standardized loadings and then also toggle to standardized loadings when viewing the values on the model. They are also found in the standardized regression weights table. As for common method bias, the threshold for differences between the CLF model and the regular model is roughly 0.200. I have a video explaining the CLF approach. I also have a new video (and plugin) to show how to add the CLF automatically.
Yikes! did you have a lot of missing data? did you have to impute a lot? Do you have reverse-coded items? I just made a new video for you. In this video, I show how you can check to see what happens when you drop one item or another from the factor. Go to my channel by clicking on my name. It is the most recent upload. Hope that helps.
Hello James, I'm following the sem series to apply for my model. so far so good, but the invarience test does not run in AMOS. I have a similar model and i want to group my model based on a nominal variable. But everytime i run the invarience test AMOS gets stuck on "reading data" do you have a solution for this? Thanks for the whole SEM series, such a good support!
For very large models, the matrix of comparisons become pretty big, and may fill up memory. If you are including more than two groups, it then becomes crazy large. Older computers would definitely struggle. If you have more than two groups, you may want to remove all but two at a time.
Thank u Professor for ur videos..they re very useful..i am just wondering if explanatory and confirmatory factor analyses are also applied to cross sectional data and time series data?
Hi James, I have been trying to use the pattern matrix builder pluggin and getting this error message- You are attempting to draw an arrow that ppoints to a variable whose variance (1) is constrained. An endogenous variable is not supposed to have a variance that is constrained.Please select an option: - do not draw he arrow -remove the constraint and draw the arrow When I choose either of the option, the variable and the error messages gets mixed up in the diagram. Please help.
That might happen if the pattern matrix you are pasting includes any single item factors, or if you are using comma notation as your decimal. Otherwise, if it still doesn't work, you might try this plugin: ruclips.net/video/z4RZhPwSV28/видео.html
@@purnimanandy4405 Then I'm not sure. Sorry about that. You'll just need to construct the model manually, as shown here: ruclips.net/video/JkZGWUUjdLg/видео.html
Hi James, I was wondering if it does matter when there are missing values in the moderating variables (i.e. gender, education). Can I just execute the invariance tests with these missing values or is it not allowed to have missing data (for the moderating variables) when executing the invariance/moderation tests? Thank you in advance!
The missing values for those variables just mean that that record in your dataset doesn't get used. AMOS takes all records listed as female in one group, all records listed as male in another group. If the record has no value for gender, then it simply doesn't get included in either group.
James Gaskin thanks for your response! Does it matter when these records are not used when executing moderation while these records ARE used in estimating the whole model?
Nienke Tiekstra If there are a ton of them, then it might introduce some sort of bias. However, if it is less than 5% of the dataset, then I wouldn't worry about it.
Hello Sir I am trying to use CFA for the model that I used for my questionnaire. But I am not getting the pattern matrix as it is supposed to be.I am getting this message in the pattern matrix box- {Pattern Matrix(a)a. Rotation failed to converge in 25 iterations. (Convergence = .003).} Please help me with this issue.
+charu saini You can try to increase the allowed number of iterations to 50. If it converges, then at least you will be able to identify the variables that are causing problems (they will have low communalities and weird loadings).
Respected Sir, Hope you are fine. Sir i am a research student of social sciences. My research model contains serial/sequential mediation (with 2 mediators in chain and 2 IVs and 1 DVs). I learnt about CFA , SEM, simple mediation and moderation from you videos on youtube channel. Thankyou so much for making such helpful videos with clear and easy communication. Hereby i am interested to know that what is the procedure to run Sequential mediation in AMOS because i am not finding any speciofic video on the concerned topic. I just want to confirm will SEM for serial mediation will be the same as we draw for simple mediation and we ll run it through bootstrapping ? Is there any additional step i need to do ? Am right? Please confirm my query. I will be grateful Best Regards
Respected Sir, I am not able to run stat tools package for group differences. It show some issues with visual basic. it is asking to debug. what shall i do. Please help me. regards Thiru Murugan
+Thiru Murugan It is probably because you have rows in your regression weights table that are constrained to 1.00 and therefore are not labeled. These will need to be removed from the table before the macro will run properly.
Dear Prof. James Gaskin, Could you please explain how your Amos can arrange the sequence of indicators based on their construct (Useful_4 to Useful_7, Joy7 to Joy1, etc) while mine messy (after two Useful indicators, then other indicators they were mixed). Thanks in advance..
It should just be in the order you entered them, unless you're using the plugin to build the CFA for you. Then it just comes in the order listed in the EFA Pattern matrix.
James Gaskin Is it ok if I use your excel tool to compare MM between groups with un-arranged indicators? just copy it to excel, and it can proceed? Thanks Sir
Dear James I have two more questions . at first, what should we do if correlations between two latent factors are high? for the secound question , is necessary to do invariance test for every data? and for what reseon?and generaly according to witch variables we should do grouping in invariance test because its hard for me to distinct . and at the end in many researchs , researchers only conduct CFA and after that they do SEM , but according to your lessons we should do EFA at first. İ'm a little confused. please guide me. REGARDS
1. If the correlation is greater than 0.900, then you might just have two factors that are really two dimensions of the same thing. You could then create a second order factor with them. Or, if that isn't the case, then you might have a problem with common method variance. 2. We only do invariance tests if we are doing multigroup moderation in the causal model. 3. I always do an EFA even if I'm using established measures. EFA reveals issues of discriminant validity much better than the CFA.
Hi Dr. James, thank u so much for sharing all these information. I followed through ur step by step series and they were incredibly helpful. 1 question though, at this CFA stage, what if I still have an item that is loading less than 0.4 under a particular latent variable? Do I delete the item or still retain it? Note that model fit indices (RMSEA, CFI, TLI, CMIN/DF) all are satisfactory even with the existence of this 1 low loading item. Thank u !
All the constructs in your example have at least 4 items. I was wondering if there was a way to get AMOS to run a CFA if one of your constructs is a single item construct? Any tips or tricks would be much appreciated.
Nope. If it has only one item, then it is not latent. The CFA is for latent factors. So, don't include it for now, but you can include it later during the structural modeling.
Really useful. The Modification Index approach is really nice and simple, but what do you do if you have missing data? AMOS won't then compute MI's will it? The alternative would be to use single imputation in SPSS, but there are quite a few papers suggesting SI isn't the best way to handle missing data. Any tips?
I usually use median imputation when dealing with Likert scale items. Or mean imputation when dealing with normally distributed continuous scale items. If more than 5% of the data is missing for a particular variable, it is hard to justify imputing. Sometimes you have to drop items or drop records in order to get usable data. Or, you could just not look at MIs, and instead look only at standardized residual covariances. But that's a bit of a pain...
James Gaskin That's really interesting. I'm a psychologist, not a statistician, and I won't pretend to be entirely familiar with the literature, but the stuff I've read, (e.g. the one below) seem to suggest that imputing leads to less biased estimates ... and I can't see any 5% rule. My previous practice was always to replace so long as there was less than 10% missing ... but I'm struggling to find any papers that recommend that. Any pointers would be HUGELY appreciated! Olinsky, A., Chen, S., & Harlow, L. (2003). The comparative efficacy of imputation methods for missing data in structural equation modeling. European Journal of Operational Research, 151(1), 53-79. doi:10.1016/S0377-2217(02)00578-7
Hi, Prof. James. I wonder how to make the pattern matrix in SPSS. Do I use correlate, then bivarate? I see your matrix is Factor 1- 7 X the latent factors. Do I have to add the columns in SPSS first? What data should I put? The mean of each latent factors? Pls advise. Thx
Dear Dr. Gaskin, thank you for the super helpful tutorials. I had a question. I was going to do a full EFA and CFA to get the results for my hypothesis, however, after the EFA, I realized I need to perform another analysis with the EFA results because CFA isnt the correct one for my research. That is because after finding the factor constructs, I need to correlate two entire factors (for example, in your specific example in the tutorial, after you complete the EFA, you proceed to find whether a correlation exists between Joy and Play or Joy and Useful). May I know what analysis you would suggest for that purpose? Thank you.
If you want to find the correlation between two latent factors, then a CFA is exactly what you want to do. If you want to find the correlation between single variables, then do a bivariate correlation in SPSS.
Dear James - You are just great! Thank you very much indeed for the wonderful videos. how can I do CFA before EFA? The video above cuts and pasts EFA data into CFA. However, I would like to just do the CFA first since I am using popular instruments for my PhD.
Mohammed Alnughaimish that is fine, as long as you still meet criteria for validity. My video called “model fit during CFA in Amos” shows how to build the CFA from scratch.
Dear James I have two questions 1- I want to apply a pre-post test for my research and I want to assess the validity and reliability of my questionnaire. how can i do that with EFA? can I do with CFA like you did in CFA section and do invariance test? 2- in in variance test procedure and configural in variance test , shouldn't we assess the model fit for both groups? because you assessed only women group after grouping
+hadi yasrebdoost Do validity just like you would with any other dataset using EFA, CFA. To do a pre-post test, you would do a t-test or anova. In the invariance test, model fit is shown for the model, rather than for each separate group. See my configural invariance video for this.
+James Gaskin could we do pre-post test like moderating effect? because if we do like the we should apply an invariance test . and for the last question should we do two efa ( for pretest and post test separately or we should do it for all of the data all together.?
+hadi yasrebdoost Usually a pre-post is for identifying differences after a treatment. The t-test or ANOVA is best for this. You can also do a homogeneity of variance test (Levene's test) to see if they are the same at the measurement level.
Hi James, I've received the following error, and wonder how can this be solved. Maybe this is because I have more than one value per parameter? e.g variable 1 has a value in factor 3 and in factor 4. System.Exception: Either the constraints on the parameters or the initial parameter values are bad. This error can occur, for example, if you request a maximum likelihood solution while specifying initial parameter values that prevent the population covariance matrix from being positive definite. Check to make sure that every endogenous variable has a residual variable (error variable) associated with it.
That is likely the cause, although it can be done, for example with an MTMM model. Try removing some of the multiple connections to see if that fixes it. If not, then try running an EFA on the data first (I always recommend EFA before CFA).
James, thank you for your videos. I watched to learn the techniques from Part 1 all the till 5. But this one uses Amos which is a software I'm not familiar with. Can I do all the things you did here on Stata?
I've never really used Stata. Most of my experience is in AMOS, SPSS, Excel, Mplus, and SmartPLS. So, I've created videos for all of those, but nothing for Stata.
Hello James, Firstly thanks for the great videos, they are really helpful in supporting me with my MSc dissertation. I have put my constructs in ascending order as you have in your tutorials, however in the pattern matrix they are jumbled? Is there anyway I can sort these in the SPSS output file?
Here's a fun pet project I've been working on: udreamed.com/. It is a dream analytics app. Here is the RUclips channel where we post a new video almost three times per week: ruclips.net/channel/UCiujxblFduQz8V4xHjMzyzQ
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!
Sir James Gaskin. Sir, how to create syntax plugin as like pattern matrix builder in AMOS and then if there have problem in error amos ; Assertion Failed Finalizer CDbfile.id=dbGraphics di CDbfile.Finalize(). How to handle it, please help me, Thanks...
@@dedylesmana The way to create a plugin is by writing the code in the AMOS plugin editor. It can be written in C# or VB.net I think. As for troubleshooting that error, it could occur for many reasons, some of which are explained here: statwiki.gaskination.com/index.php?title=Plugins#Troubleshooting
@@Gaskination thanks sir.
@@Gaskination Sir, I want to questions, when pattern matrix builder of CFA model builder but how does not appropriate with factor it in the models
@@dedylesmana I don't think I understand your question. Here is a troubleshooting guide that might help: statwiki.gaskination.com/index.php?title=Plugins#Troubleshooting
Hi James, just wanted to thank you for all your AMOS videos. They've been most helpful in helping me to write an article using CFA. Thankyou!
CFA serves many of the same purposes, but not all. EFA is good for unconstrained factor analysis, in which you are not assigning items to factors. CFA assigns items to factors, and allows for more robust tests of discriminant validity. You can also test common method bias and group invariance better during the CFA.
Thank you for your videos and the applications you developed. Thanks to your valuable help I was able to finish my PhD!
No, I do not have a lot missing data and not using reverse-coded items. I would check with your video soon and see what happen next. Btw, thank you very much James for your help. Really appreciate much your help with all these videos. Excellent work!
2) After calculating the model and showing the numbers, you say the value should be above .7 but I have a few below. However I think the ones lower are where factors where 2 factors in EFA loaded together and I wanted CFA with the same model before trying my pre EFA model or even deleting some items to solve the cross loading.
Thanks a lot for your time!
Estimation issues. Might have negative error variances. Check standard errors for anomalies. Watch my video on "iteration limit reached in amos" to get some ideas.
Hello James entire SEM series is really helpful for me . I am badly stuck up with one problem for a particular model where its showing Iteration limit reached . I kept Iteration limit as 50 . But the results for fit indices are satisfactory .
Look at the Notes for the Model section of the output to see if it gives any clues. You can also watch my "iteration limit reached in amos" video for some additional clues.
Hi Sanjukta,
RUclips isn't letting me reply to your comment directly, so I will reply here. I have a video that discusses "iteration limit reached in amos" Watch that for some ideas. I also go over it in my bootcamp videos when I talk about common method bias.
I have a question regarding the Experience high and low and frequency, how you did calculate them in spss at first and also how i can identify mine. Btw, thanks a lot for your great and helpful videos.
Kholoud Elaimi I calculated them using the median. However, the easiest way is to use the Rank feature in the Transform menu in SPSS.
James Gaskin Thanks a lot for your quick reply, what if i have them as scale multiple answers, how can i do that if you dont mind???
Kholoud Elaimi I don't understand. Please give an example.
James Gaskin OK, my moderators are for example technology level, and frequency usage of that technology, I have those in my SPSS data as multiple answer questions since they are Likert scale 1 to 7 style. sp how i can group them first to 0 (1-3) low, then 1 (4-7) high, then i can chose them in my Amos.
Also in the part where you have CLP the common bias, u had a negative results from joy but in my case i had them with some of my variables with CLP, so how can i fix that, because i tried your moving the parameter way, but still, do i delete those negative loading factors ??? or what? I'm so grateful for your generous help. Thanks a lot.
Hi
I installed the plugin but when I paste the pattern matrix it is really messy therefore the diagram is not correct as well.
I would appreciate if you help me to solve this problem .
thank you
I would recommend resolving the EFA first, then doing a CFA. If the matrix is messy from crossloadings, then just suppress loadings less than 0.300. The plugin also is known to fail if you are using comma notation instead of decimal notation (e.g., 0,05 instead of 0.05) or if you are using variable labels instead of variable names in the EFA.
how to get a pattern model build up? could you please help?
It is sometimes also called the component matrix or the factor rotation matrix.
No. Since word of mouth is not latent, it does not belong in the CFA. The CFA is only for latent variables. You can add word of mouth when you start testing a causal model.
Hi James, the Excel file is very helpful, however, when I am writing the thesis, how do I make justification of using this file as I think the examiner will not accept it if I say I downloaded from the internet. Thanks James.
You can cite it or use one of the updated plugins and cite that. How to cite the excel file and the plugins is shown on the homepage of statwiki. Thanks!
Yes. Do Low:High, Low:Med, Med:High. Or do Low:MedHigh, Med:LowHigh, High:LowMed.
Wow! I totally forgot that one! Yikes. Thanks for the note. I'll make it right now to complete the CFA training.
I've never seen an error with that specific wording, but it sounds like a variable with that exact name is not appearing in the specific dataset you linked in AMOS. You might have linked the wrong dataset or spelled the variable name incorrectly.
I would not drop the factors so readily. I would first fight for their survival. See if there are data normality issues, outliers, convergent validity issues, etc. Make sure they work in the EFA after that. Then, after all of that, test for reliability. If they are close to 0.700, move forward. Drop as little as possible.
HI James!
So, VERY NEW to these procedures. I don't understand where the pattern matrix comes from if SPSS only computes EFA or PCA. Thanks for providing feedback on this question that is probably silly.
I have to say.. Thank you very much!! You saved my life!!
They should average out to above 0.70, so it is okay to have some below if, on average, they are above 0.700.
Sorry for the typos, I am writing in a hurry from my mobile phone.
For analyzing the factor structure of higher order constructs, you should watch my video about 2nd order factors in AMOS.
Hi James, thanks a lot. I have a question please, What if the results of EFA were different from those of theory? we conduct CFA basing on which of them? Let's assume that we have a concept "attitude" which theoretically has three dimensions (cognitive, affective & behavioral), but the results of EFA shows only one factor. In this case should we ignore EFA & treat Attitude as a second order factor withe three dimensions? Or we should respect the EFA results & treat it as a single one order factor? I have such problem in my research & I couldn't figure out what to do.
+Mohamed Hasan 2nd order factors sometimes need to be run as their own EFA. This is what I usually do. Then in the CFA, model it as 2nd order.
For large models (sample size greater than 200 and/or variables greater than 20) it is very difficult to achieve a non-significant value.
Thanks a lot for your prompt reply, James!
God bless you Prof. James . Good videos , good job!
Hi James. Need to ask.. Say that I have 5 factors in EFA and turned out that only the first 2 factors are reliable and the other 3 factors need to be deleted (Cronbach's alpha
Thanks for your fantastic videos, not everybody is willing to share these knowledge for free. Like what you always say, SMILE :-) , That's what I do when I watch your videos.
One question, is there any way to use the stat tool package for a variable with three group, like low, medium and high?
great work James! can you do a video on how to use sem for validating a likert type scale ??
James, can you talk about the rationale behind the use of modification indices in a confirmatory framework. Specifically, why look at the residual covariances? Or, the larger question, are we actually doing theory driven work if we are relying on data output to improve model fit? Thoughts?
Model fit is all about accounting for all the major covariances among the set of variables in your model. If they are accounted for and you still have strong loadings and good discriminant validity, then we are satisfied that we can rely on the measurement model. Thus, addressing modification indices and standardized residual covariances is not warping our model or fishing around. It is simply "controlling" for major covariances we didn't theorize, in order to validate the covariances we did theorize. I hope that helps.
great work James!
when I did EFA I received this error Warnings
Text: Q63_Comm
An undefined variable name, or a scratch or system variable was specified in a variable list which accepts only standard variables. Check spelling and verify the existence of this variable.
This command not executed.
I have 10 variables with 85 items. would you please help me and explain how can I solve this error?
Hello Mr. Gaskin, I have been trying to use the stats tool package suggested by you, however after loading the data onto the downloaded excel sheet, it does't work and displays a message saying 'debug'.
How can I avail the correct version of the software?
Thank You.
Shivangi Sanganeria Make sure to look at the caveats and assumptions. If it still doesn't work, then email it to me and I'll take a look. james.gaskin@byu.edu
But in the Stats Tool Package there are only two boxes for groups to paste the tables. I think we have to add something in the Visual Basic which I am not expert in it.
Did all that, checked for the normality issues, outliers, convergent validity issues and all that.. They surely did work in EFA but still are not reliable (Not even close to .7, ranges from 0.5 - 0.6)
Thanks a lot for yet another great video. I am walkung step by step with you and everything is looking fine. I have two quick questions if I may:
1) The modification indices area only have some factors under the regression part. Do I need to remedy these? All of them dissappeared after I created groups for invariances for age and experience even when options still show them as on.
When analysing the common method bias; if you have negative variables/values, does this mean the same as if they are low i.e. that we have differences in standard regression weights when adding the common latent variable or does negative variables/values mean that these are not acceptable at all? I have tried to move the regression weight to other indicators however, this doesnt seem to work on the results either.
Thank you! Saved me a lot of time on my thesis.
Good job. I really like your videos
Great teacher!
Hi James,
Many thanks for such wonderful video series.
May I ask about a problem I am encountering. When I run CFA analysis, an error message popped up saying that "Either the constraints on the parameters or the initial parameter values are bad. This error can occur, for example, if you request a maximum likelihood solution while specifying initial parameter values that prevent the population covariance matrix from being positive definite. Check to make sure that every endogenous variable has a residual variable (error variable) associated with it". I am using your data file, go step by step following your videos. Could you please help advise how to fix the problem?
Great thanks James
Huyen
Hello James, I have a problem performing CFA. Every time when i run the analysis, Standardized Regression Weights are not calculated. What can be the problem?
Hi James, i solved the problem,I constrained the error variance to a fixed value. previously, it was fixed at zero and now i made it to be 1. Now, it shows the standardized values. Thanks anyway. Your videos are very useful. all the best
Hi James, Thanks for the video, which is quite helpful. ONE question: what do you think of the interpretation of residual correlation in CFA? Some scholars said in managment practice, error term corrleation is not recommended or welcomed by good journals; some others said it may indicate content overlapping in items. I find it hard to explain when validating a scale. Thanks.
Thank you, professor Gaskin! I wasn't aware of that video! It is very helpful!
I have just one question. Namely I have noticed that in the video, in your Notes for Model part of the output, the probability level is 0.433, while in your SEM videos it always is 0.000. I know it is related to the sample size, but is there a rule of thumb to know when should I accept significant or nonsignificant chi-square?
If you will be moderating in your path model, then you can leave these unaddressed. I would then report the model fit for the moderated model.
Hi James.
I have a question, please. I'd like know the source (article, book, or chapter) where I can find the recommendation about the use of the comparison of critical ratios criteria in the group comparison/differences (11:35 to 11:43). I'll use this criterion in my analysis.
Thanks.
It is not very well supported in the literature. Here is an updated video with more support (via chi-square difference testing): ruclips.net/video/Pakg3PlppuY/видео.html
Hi James!!
I am trying to run my first SEM with AMOS for my thesis and I am having some trouble to proceed with it...The first problem is that I have a model with 13 latent variables and 52 observed variables, and I do not know if I have to make the first step that you make in this video with all of them and covariating all 13 constructs (as the graphic seems kind of crazy and it doesnt let me run the analys...coul you give me a hand with that?
On the other hand, I get these kinds of errors when I try to run the complete model (including unidirection arrows from the independent to the dependent variabes of my model:
- The covariance matrix is not positive definite: whta does this mean and how can I solve it?
- The model is probably unidentified. In order to achieve identifiability, it will probably be necessary to impose 1 additional constraint ( I have fixed 1 in all arrows from errors to observed variables and in 1 of the arrows that go from each latent variable to its observed variables...what more constraints can I make?
- an error occured while attemting to fit the model either the constraints on the initial parameters or the iniciatl parameter values are bad (...). What does it mena and what can I do to be able to run the analysis withouth this error message?
I AM QUITE DESPERATE AS I AM RUNNIGN OUT OF TIME WITH MY THESIS, AND SOI I WOULD REALLY APPRECIATE YOUR HELP
I love you videos by the way!! They are super useful and clear for beginners lke me ;)
Elisa
Elisa López Sorry for the delay. I was hosting an SEM Boot Camp last week and then yesterday was my wife’s birthday. Today is my first day back… To answer your questions:
1. Yes, all the latent factors should be covaried. The reason it won't run might be due to sample size. You have a fairly complex model. If the sample size is not greater than about 200, you will have trouble running it. If you do have enough sample size, then it could just be something about the way it is drawn. Try recreating it with my EFA-CFA plugin available on my wiki. Here is a video about how to get it: ruclips.net/video/sLtMOFcojZY/видео.html
2. The errors you have are also likely due to an error in drawing the model. Again, I recommend the plugin as it removes the possibility of creating such errors. If you still cannot get it to work, then you can email me your .sav and .amw files and I can look at it.
Hello,
I want to ask about the modification indices. I have two observed variables that seem to have high scores, sort of like in the video. You mentioned that covaried them, but I don't understand the reason for doing so. What does this do and how can be it explained?
It accounts for the correlation between their errors. Model fit is all about accounting for major covariances (or correlations). So, if there is a major correlation between two variables, or errors, then you can account for that with a covariance arrow. When we do this with error terms, this means we want to account for their correlation without explicitly theorizing about their relationship.
Dear James, I proposed the following model (regarding service recovery):
(interactional justice, procedural justice & disutributive justice) lead to (transaction specific satisfaction & overal firm satisfaction) leading to (repeat purchasing & word of mouth).
All 7 variables are tested with 3 or 4 items (in a survey). However, word of mouth is tested using only 1 item. When performing the CFA (step 5.1), should I also include word of mouth in the model, and if so, how can I do that?
Hello, I currently have SPSS Amos version 26, I made the indications of the file "FIX Syntax Error with Pattern matrix 24 plugin.doc" to copy the file "PatternMatrixBuilderAMOSv24.dll", but I get an error, what should I do?
Unfortunately, you might have to manually build the model as shown in this video: ruclips.net/video/JkZGWUUjdLg/видео.html
Dear Prof Gaskin, I run a CFA but all the regressions are unidentified. The model has 2 latent variables with 5 indicators for one of them and 6 for the other. I run an EFA before the CFA and get rid of cross loadings. The reliability is also in the accepted rates. The bivariate correlation between the two latent variables is 0.578. The sample size is 216 cases. Could you help finding the problem? Thanks
bayan khalifa Check the notes for the model in the output to see if there are any clues there. For example, it might say you are missing a constraint, or that you reached the iteration limit.
Hi James and thank you for this tutorial on SEM. It is extremely useful for me. I'm developing a model and I got these results for the default model in AMOS:
Result (Default model)
Minimum was achieved
Chi-square = 4,252
Degrees of freedom = 4
Probability level = ,373
How do I report this? Is this a good model to report or does it have to have a probability level < ,05? I'm really confused with this probability level.
Other results for this default model are CMIN/DF 1,063, CFI ,999, PCLOSE ,881 and RMSEA ,010. If I understand correctly, all these results show a good model?
Thank you so much!
Hi James!
Thanks for your videos, they have been very useful for me.
I have a question, my Amos Output don't have "Corelations" and "Covariances" under Estimates - Scalars. Do you know what can be the problem?
+Seba CEE Make sure you have checked the box for standardized estimates in the analysis properties window.
Thank you!
Hi Dr Gaskin,
After completing the process of screening of data and EFA, when run the CFA initial model, I got this error 'GROUP 1, check for incomplete data', I followed your each step and used the same data as you did, but don't know what happened? please help.
This just means that you have some missing data. So maybe you need to check to see if you have variables in your model that have blank cells in SPSS.
Hi James, am following your videos and I am just wondering why one would do a CFA after an EFA on the same dataset? I have seen papers published with just the EFA done to show factor structure. I thought the main reason to do the CFA would be for new samples?
Hello James,
I have been following your videos and have found them amazingly helpful! But I ran into a little snag when trying to paste my Pattern Matrix from SPSS into the AMOS Pattern Matrix Input plugin window - I can select and copy the Pattern Matrix, but I can't paste it into the pattern matrix builder window in AMOS (when I right mouse click the 'paste' option is grey/unavailable & using paste short-cut key does not work either). I believe that I have correctly installed the plugin in AMOS (the window opens and all looks okay) & there is no problem with copying the pattern matrix (i.e., I can paste it into a word doc), but it's just not allowing me to paste into the Pattern Matrix Input window. Have you come across this problem before, and/or do you have any possible solutions that I might try to fix it?
Many thanks, Kim
Kim Woodward I haven't come across this before. The only reason AMOS would not let you paste would be because it does not recognize that anything is on the clipboard. Make sure to copy the pattern matrix before opening the plugin. If this is what you did, then try copying after opening the plugin. If this still doesn't work, then I'm not sure. I suppose you'll have to do it the good old fashioned way... Here is a video where I show how to build it from scratch: ruclips.net/video/JkZGWUUjdLg/видео.html
Hi James - you're saving so many PhDs, I am sure :) For some reason, in my output for the invariance test, the 'Pairwise Parameter Comparisons' isn't there? I've gone through the video a couple of times to see if I've missed checking something, but cannot see that I have. Anything else I can check? TIA (thanks in advance)
+Jennifer Barhorst It should show up if you have multiple groups assigned, if you have assigned those groups data, and if you have checked the box for critical ratios for differences.
Hi James,
Thank you so much for this video.
I run my model and I get the results. The results showed that one out of 5 factors in my questionnaire correlated negatively with another factor. Is that mean the questionnaire not valid? Is there such way to sort out this issue? Thanks
Sorry to bother you again. At first, my model showed a bad fit. The fit became acceptable after some modification. However, after adding the CLF the model fit has increased again. If I have to report model fit in a paper, at what point in time do I measure fit?
Dear James
İ have done a CFA and eliminate some factors and draw some correlations according to your youtube lessons and achive acceptable goodness of fit but validity and relibility is very low. what should İ do.? should İ only say that the model is unvalidated or İ have other ways?
REGARDS
You can't move forward with it, or else all your next analyses will be invalid.
Dear Prof. James Gaskin. Is it ok if I use your excel tool to compare MM between groups with un-arranged indicators? just copy it to excel, and it can proceed? Thanks Prof
I don't think I understand the question. However, my excel tools are designed for AMOS output only. They are very NOT robust. The tools expect the exact structure of the AMOS output.
Hi,
Is there anywhere in this series or your other videos where you discuss how to use observed variables of different measurement types as indicators of the same latent variable?
Not that I can think of. If you mean they are of different scales, then that is fine. If you mean they are of different dimensions of the factor, then you will want to use formative modeling.
Hi James!,
Can you tell me is your plugin works on new version of Amos (v24), because I have got a problem with loading it. I will be grateful for your answer :)
+Natalia Przybylska It does not. AMOS 24 changed the way it uses plugins... I'm working on making it compatible. Hopefully sometime this month.
Hello, professor Gaskin. Thank you for your videos and an extremely helpful explanations and examples.
I have only one issue when performing the common method bias. Namely, I get the error "A sample of parameter values was inadmissible" and my new data set with composites is not imputed. Do you have an advice on that? Many thanks.
Hi James,
I'm following your videos step by step. My final table in the GroupDifferenceTab has all the columns equals to the ones shown in the video apart from the z-score column. I don't understand why all the variables have a sig. value (0.000). What could have happened?
If all the variables have sig values of 0.00, then that just means they are all significant indicators of their factors.
Hi James my name is Emile I am a PhD students in South Africa. I find your video very interesting. I have been asked to analyses the "Factor structure of PsyCap and OCB, internal consistency
of each of the sub scales, equivalent unbais measuring across
different ethnic groups".Psycap and OCB are two constructs with sub construct; I was wondering if this is the right video to watch could you give me some advises on how to proceed?
Thank you
Thank you. This was very helpful. I only have one question in regards to the video. As you showed; you need the pattern matrix however, when I did my factor analysis in SPSS by making use of extraction method: principal component analysis and rotation method: varimax, I did not recieve a pattern matrix. Does this not allow me to do an CFA or can I use something other than the pattern matrix?
Hello,I was following your example trying to use one of my previous pattern matrix, however, I cannot get the file to upload in AMOS. What am I doing wrong?
It must be copied directly from the spv file. It cannot come from Word or Excel.
Hi James,
I am comparing samples from two different countries.
Please guide me in the following isues:
Do i need to perform CFA on different data sets separately or i can combine them?
While running the structured model how do I compare two different populations?
Is it Fine, If factor loading are more than 1 among variables and constructs in CFA results ?
1. ruclips.net/video/sbTl-icEyD8/видео.html
2. ruclips.net/video/w5ikoIgTIc0/видео.html
3. ruclips.net/video/Vx24KFf-rAo/видео.html
Express fit without the CLF.
Dear James
İ want to apply a distincted model (İ Have taken it from another research) and research the validaty and reliability of that on a new sample . but İ dont know what İ should do? İ think İ should do efa before cfa but the problem is that the the results of efa is very defferent with majour model, please reply me if İ should do efa at firs or cfa ?
I always do EFA first if I have new data, even if the constructs are all well-established in the literature. The EFA reveals discriminant validity issues much better than the CFA can.
Hello James, I used to Pattern Matrix Builder to paste my EFA but the loadings were running, the display didn't come out as yours did. How I can make it look like yours?
Daisy G I'm not sure what you mean by "running". Sometimes the plugin doesn't work when you use variable labels instead of variable names in SPSS. It also has been known to fail when using commas instead of decimals (as done in many European countries). If all else fails, you can try to build it by hand as shown in this video: ruclips.net/video/JkZGWUUjdLg/видео.html
James Gaskin Thank you so much. I still can't figure out why. As I cut and paste the EFA into pattern matrix builder, the arrow of the error values are not in the right place. And the whole model doesn't space like yours in the video. I am gonna try changing that variable labels and see if that works. Thank you.
Hi Prof.James, when I tried to run my CFA model. I kept getting this error message that "the variable, "nameofvariable" is represented by a rectangle in the path diagram, but it is not an observed variable". I hope that you would be able to enlight me on What could have been the cause of this error? I am not sure how to deal with this as I am quite sure that the variable is an observed variable.
thanks.
+Law jiawern This can happen if it is spelled or spaced incorrectly, or if you accidentally uploaded an earlier version of the dataset or the wrong dataset. A way to prevent this is to simply go to view, variables in the dataset, and then drag the variable into the rectangle.
Noted, thanks!
oh my god i think i love you! Thank you so much for all your help :)
Dear James,
I am following this video with a similar model, however I have some problems. When I run my CFA, most values between the factors and items are higher than 1. Is that possible? Aren´t they supposed to be between -1 and 1?
Futher I above my facotors there are values between .99 and 2.35, which I cannot spot in your model. Is that an additional setting to be taken before I run the model?
Thank you very much for your help and the great video series in general! :-)
One further question concerning the Common method bias. Is there a general threshold?
The factor loadings should be between -1/+1, but my guess is that you are looking at UNstandardized loadings. You need to check the box for standardized loadings and then also toggle to standardized loadings when viewing the values on the model. They are also found in the standardized regression weights table. As for common method bias, the threshold for differences between the CLF model and the regular model is roughly 0.200. I have a video explaining the CLF approach. I also have a new video (and plugin) to show how to add the CLF automatically.
Thanks James! I will look for that!
Yikes! did you have a lot of missing data? did you have to impute a lot? Do you have reverse-coded items? I just made a new video for you. In this video, I show how you can check to see what happens when you drop one item or another from the factor. Go to my channel by clicking on my name. It is the most recent upload. Hope that helps.
Hello James,
I'm following the sem series to apply for my model. so far so good, but the invarience test does not run in AMOS. I have a similar model and i want to group my model based on a nominal variable. But everytime i run the invarience test AMOS gets stuck on "reading data" do you have a solution for this?
Thanks for the whole SEM series, such a good support!
For very large models, the matrix of comparisons become pretty big, and may fill up memory. If you are including more than two groups, it then becomes crazy large. Older computers would definitely struggle. If you have more than two groups, you may want to remove all but two at a time.
HI James, Thanks for the wonderful video. I am trying to do CFA for 11 variables. can I do it at one go?
Thank u Professor for ur videos..they re very useful..i am just wondering if explanatory and confirmatory factor analyses are also applied to cross sectional data and time series data?
Hi James, I have been trying to use the pattern matrix builder pluggin and getting this error message- You are attempting to draw an arrow that ppoints to a variable whose variance (1) is constrained. An endogenous variable is not supposed to have a variance that is constrained.Please select an option:
- do not draw he arrow
-remove the constraint and draw the arrow
When I choose either of the option, the variable and the error messages gets mixed up in the diagram.
Please help.
That might happen if the pattern matrix you are pasting includes any single item factors, or if you are using comma notation as your decimal. Otherwise, if it still doesn't work, you might try this plugin: ruclips.net/video/z4RZhPwSV28/видео.html
@@Gaskination I tried but this pluggin not working either:( the program just hangs.
@@purnimanandy4405 Then I'm not sure. Sorry about that. You'll just need to construct the model manually, as shown here: ruclips.net/video/JkZGWUUjdLg/видео.html
Hi James, I was wondering if it does matter when there are missing values in the moderating variables (i.e. gender, education). Can I just execute the invariance tests with these missing values or is it not allowed to have missing data (for the moderating variables) when executing the invariance/moderation tests? Thank you in advance!
The missing values for those variables just mean that that record in your dataset doesn't get used. AMOS takes all records listed as female in one group, all records listed as male in another group. If the record has no value for gender, then it simply doesn't get included in either group.
James Gaskin thanks for your response! Does it matter when these records are not used when executing moderation while these records ARE used in estimating the whole model?
Nienke Tiekstra If there are a ton of them, then it might introduce some sort of bias. However, if it is less than 5% of the dataset, then I wouldn't worry about it.
Hello Sir
I am trying to use CFA for the model that I used for my questionnaire. But I am not getting the pattern matrix as it is supposed to be.I am getting this message in the pattern matrix box- {Pattern Matrix(a)a. Rotation failed to converge in 25 iterations. (Convergence = .003).}
Please help me with this issue.
+charu saini You can try to increase the allowed number of iterations to 50. If it converges, then at least you will be able to identify the variables that are causing problems (they will have low communalities and weird loadings).
Respected Sir,
Hope you are fine. Sir i am a research student of social sciences. My research model contains serial/sequential mediation (with 2 mediators in chain and 2 IVs and 1 DVs). I learnt about CFA , SEM, simple mediation and moderation from you videos on youtube channel. Thankyou so much for making such helpful videos with clear and easy communication.
Hereby i am interested to know that what is the procedure to run Sequential mediation in AMOS because i am not finding any speciofic video on the concerned topic. I just want to confirm will SEM for serial mediation will be the same as we draw for simple mediation and we ll run it through bootstrapping ? Is there any additional step i need to do ? Am right?
Please confirm my query. I will be grateful
Best Regards
+hafsa noreen That is correct. Just run it with bootstrap and you'll be able to see all the indirect effects.
Correct me if I'm wrong but the patter martix that was copied here was from the EFA?
yes
Respected Sir,
I am not able to run stat tools package for group differences. It show some issues with visual basic. it is asking to debug. what shall i do. Please help me.
regards
Thiru Murugan
+Thiru Murugan It is probably because you have rows in your regression weights table that are constrained to 1.00 and therefore are not labeled. These will need to be removed from the table before the macro will run properly.
Dear Prof. James Gaskin,
Could you please explain how your Amos can arrange the sequence of indicators based on their construct (Useful_4 to Useful_7, Joy7 to Joy1, etc) while mine messy (after two Useful indicators, then other indicators they were mixed). Thanks in advance..
It should just be in the order you entered them, unless you're using the plugin to build the CFA for you. Then it just comes in the order listed in the EFA Pattern matrix.
James Gaskin Is it ok if I use your excel tool to compare MM between groups with un-arranged indicators? just copy it to excel, and it can proceed? Thanks Sir
Dear James
I have two more questions . at first, what should we do if correlations between two latent factors are high? for the secound question , is necessary to do invariance test for every data? and for what reseon?and generaly according to witch variables we should do grouping in invariance test because its hard for me to distinct . and at the end in many researchs , researchers only conduct CFA and after that they do SEM , but according to your lessons we should do EFA at first. İ'm a little confused. please guide me.
REGARDS
1. If the correlation is greater than 0.900, then you might just have two factors that are really two dimensions of the same thing. You could then create a second order factor with them. Or, if that isn't the case, then you might have a problem with common method variance.
2. We only do invariance tests if we are doing multigroup moderation in the causal model.
3. I always do an EFA even if I'm using established measures. EFA reveals issues of discriminant validity much better than the CFA.
Hi Dr. James, thank u so much for sharing all these information. I followed through ur step by step series and they were incredibly helpful. 1 question though, at this CFA stage, what if I still have an item that is loading less than 0.4 under a particular latent variable? Do I delete the item or still retain it? Note that model fit indices (RMSEA, CFI, TLI, CMIN/DF) all are satisfactory even with the existence of this 1 low loading item.
Thank u !
If the AVE and CR are good, and discriminant validity is established, then you can keep the item.
All the constructs in your example have at least 4 items. I was wondering if there was a way to get AMOS to run a CFA if one of your constructs is a single item construct? Any tips or tricks would be much appreciated.
Nope. If it has only one item, then it is not latent. The CFA is for latent factors. So, don't include it for now, but you can include it later during the structural modeling.
Really useful. The Modification Index approach is really nice and simple, but what do you do if you have missing data? AMOS won't then compute MI's will it? The alternative would be to use single imputation in SPSS, but there are quite a few papers suggesting SI isn't the best way to handle missing data.
Any tips?
I usually use median imputation when dealing with Likert scale items. Or mean imputation when dealing with normally distributed continuous scale items. If more than 5% of the data is missing for a particular variable, it is hard to justify imputing. Sometimes you have to drop items or drop records in order to get usable data. Or, you could just not look at MIs, and instead look only at standardized residual covariances. But that's a bit of a pain...
James Gaskin That's really interesting. I'm a psychologist, not a statistician, and I won't pretend to be entirely familiar with the literature, but the stuff I've read, (e.g. the one below) seem to suggest that imputing leads to less biased estimates ... and I can't see any 5% rule. My previous practice was always to replace so long as there was less than 10% missing ... but I'm struggling to find any papers that recommend that. Any pointers would be HUGELY appreciated!
Olinsky, A., Chen, S., & Harlow, L. (2003). The comparative efficacy of imputation methods for missing data in structural equation modeling. European Journal of Operational Research, 151(1), 53-79. doi:10.1016/S0377-2217(02)00578-7
Hi, Prof. James. I wonder how to make the pattern matrix in SPSS. Do I use correlate, then bivarate? I see your matrix is Factor 1- 7 X the latent factors. Do I have to add the columns in SPSS first? What data should I put? The mean of each latent factors? Pls advise. Thx
Do an exploratory factor analysis. I have several videos on the EFA. You go to Analyze, Dimension Reduction, Factor Analysis.
James Gaskin Thanks. Will try.
Dear Dr. Gaskin, thank you for the super helpful tutorials. I had a question. I was going to do a full EFA and CFA to get the results for my hypothesis, however, after the EFA, I realized I need to perform another analysis with the EFA results because CFA isnt the correct one for my research. That is because after finding the factor constructs, I need to correlate two entire factors (for example, in your specific example in the tutorial, after you complete the EFA, you proceed to find whether a correlation exists between Joy and Play or Joy and Useful). May I know what analysis you would suggest for that purpose? Thank you.
If you want to find the correlation between two latent factors, then a CFA is exactly what you want to do. If you want to find the correlation between single variables, then do a bivariate correlation in SPSS.
Thank you very much :)
Dear James - You are just great! Thank you very much indeed for the wonderful videos. how can I do CFA before EFA? The video above cuts and pasts EFA data into CFA. However, I would like to just do the CFA first since I am using popular instruments for my PhD.
Mohammed Alnughaimish that is fine, as long as you still meet criteria for validity. My video called “model fit during CFA in Amos” shows how to build the CFA from scratch.
Thank you, it's great and informative.
Dear James
I have two questions
1- I want to apply a pre-post test for my research and I want to assess the validity and reliability of my questionnaire. how can i do that with EFA? can I do with CFA like you did in CFA section and do invariance test?
2- in in variance test procedure and configural in variance test , shouldn't we assess the model fit for both groups? because you assessed only women group after grouping
+hadi yasrebdoost Do validity just like you would with any other dataset using EFA, CFA. To do a pre-post test, you would do a t-test or anova. In the invariance test, model fit is shown for the model, rather than for each separate group. See my configural invariance video for this.
+James Gaskin could we do pre-post test like moderating effect? because if we do like the we should apply an invariance test . and for the last question should we do two efa ( for pretest and post test separately or we should do it for all of the data all together.?
+hadi yasrebdoost Usually a pre-post is for identifying differences after a treatment. The t-test or ANOVA is best for this. You can also do a homogeneity of variance test (Levene's test) to see if they are the same at the measurement level.
Hi James, I've received the following error, and wonder how can this be solved. Maybe this is because I have more than one value per parameter? e.g variable 1 has a value in factor 3 and in factor 4.
System.Exception: Either the constraints on the parameters or the initial parameter values are bad. This error can occur, for example, if you request a maximum likelihood solution while specifying initial parameter values that prevent the population covariance matrix from being positive definite.
Check to make sure that every endogenous variable has a residual variable (error variable) associated with it.
That is likely the cause, although it can be done, for example with an MTMM model. Try removing some of the multiple connections to see if that fixes it. If not, then try running an EFA on the data first (I always recommend EFA before CFA).
James, thank you for your videos. I watched to learn the techniques from Part 1 all the till 5. But this one uses Amos which is a software I'm not familiar with. Can I do all the things you did here on Stata?
I've never really used Stata. Most of my experience is in AMOS, SPSS, Excel, Mplus, and SmartPLS. So, I've created videos for all of those, but nothing for Stata.
Hello James,
Firstly thanks for the great videos, they are really helpful in supporting me with my MSc dissertation.
I have put my constructs in ascending order as you have in your tutorials, however in the pattern matrix they are jumbled? Is there anyway I can sort these in the SPSS output file?
Rula Danias Just don't check the box for ascending order.