Model fit during a Confirmatory Factor Analysis (CFA) in AMOS

[stataguy]

why can't we covarite the error terms of different factors? what if there is methodological issue, for example, similar wording for survey questions? can you suggest a reference for this, please? thanks.

[Gaskination]

@stataguy The detailed answer is yes, you can covary any errors if there is a good reason for systematic correlation of residuals. However, if the correlation is due to a causal relationship (rather than similar wording - thus systematically related), then you should not covary them. In the video I try to keep it simple. Hope this helps. I can't think of a reference off the top of my head.

[ATarhini]

Hello, Do i have to inculde the moderating variables such as Culture constructs (like power distance, uncertainty avoidance...) in the CFA, m measuring the culture using 7 likert scale. cheers

[Gaskination]

@ATarhini

I don't know if there is a right or wrong answer to this. I would include in the CFA all latent variables that I intended on using in my model. This would establish that they are distinct constructs. Moderators, in particular, should not be strongly correlated with the other variables in the model, so I would include them just to make sure they meet this criteria.

[ATarhini]

@Gaskination Many thanks !!

[tony287410]

@Gaskination

Thank you for explaining complex things in simple way. my queries are: what is the importance of model fit in CFA, can we interpret that the model established in model fit is statistically approved . What will be the next step to check cause and affect relationships among the latent constructs can you provide link from your videos. Many thanks

[Gaskination]

@tony287410 We do model fit during the CFA in order to establish that the factor structure we have come up with is a satisfactory one. The next thing you would do to establish causality is to start drawing regression lines between the latent constructs. Check out my video "From Measurement Model to Structural Model in AMOS".

[ATarhini]

Hi, Thaks for the video, just want to know if its better to delete the problomatic variable or just do the covariane with other variable within the same construct.

[Gaskination]

@ATarhini It depends on if it is adding any value. If it is truly contributing to the factor, then keep it, if it is not, then drop it. covarying the error terms (which is what I assume you are talking about) is just one way to keep it without causing issues.

[ATarhini]

@Gaskination But would it cause issues when doing the construct validity later on ? as the CR for those two varaibles are < .41 and .45 respectively. knowing that I have 6 variables to measure this construct. deleting two will not be a problem i think

[Gaskination]

@ATarhini If they are reflective (interchangeable) and you have six of them, I would probably just remove one.

[vyeniaras]

Hi, thanks for the video! Can you just remove variables because their loadings are low? If so what is the criterion? Thank you!

[Gaskination]

@vyeniaras

This tutorial is meant to be a mechanical demonstration. So, trimming is really more subjective than I make it out to be. To meet the criterion for convergent validity (AVE>0.50, CR>0.70) then loadings on a single factor should at least average out to > 0.70. However, if you are working with established measures, and you're not worried about the validities, you can probably accept loadings as low as 0.30 (several references for this). Accepting low loadings may cause other problems.

[Gaskination]

@vyeniaras

Additionally, you may not want to let go of a certain item because it is crucial to your construct (however, this should not be an issue when using interchangeable items, as should be done for reflective constructs).

[urownsherry]

hi ther i need help wd this cfa thingy, when i lod my item it gives me a series of erros can u help plzzzzzzzzzzzzzzzzzzzzzzz

[Gaskination]

@urownsherry

What are the errors? you can email me directly at james. eric. gaskin@ gmail. com

[Boyana1981]

Thanks for the very useful instruction! I have a question: I'm strugging with a second order factor analysis, I'm doing it in Lisrel. When I try to connect my first and second order latent constructs I get the error message "Path from Ksi-variable to ksi-variable is not allowed". Do you know how to build a correct second order model? Thanks a lot!

[Gaskination]

@Boyana1981

I know how to do it in AMOS, but I've never used Lisrel.

[pretorianos76]

Very helpfull.Thanks a lot.MICHAEL

[xademe]

such a complete, concise, and useful tutorial.

[AtyDeh]

Thank you for the excellent explanation, i have a question, if we regress all above factors to an observed variable (say performance) which has error term, how we can write an equation for performance with regard to all factors and their covariance relation as well as error terms?

Thank you,

[Gaskination]

@AtyDeh

As far as I understand, you could write an equation for each relationship, but not for all the relationships in a single equation. AMOS does not produce this equation for you. You would simply write each equation as you would write any other linear regression equation.

[farispt]

Thanks for the great video. My question is, Can we covary the item errors in CFA without theoritical bases?

[Gaskination]

@farispt

The theoretical basis is that they are reflective and interchangeable items, which means that they were probably worded very similarly, which means that they probably have a systematically related error (rather than a causal one). So, yes, you can covary the error terms as long as they are within the same factor.

[AtyDeh]

this video just working upto 50 sec, could you please chek it James? I am able to watch all other videos from you just i found difficulty with this :(

[Gaskination]

@AtyDeh

I just checked again. It works great for me, all the way through. Sorry! Maybe try from a different computer.

James

[AtyDeh]

Hi,

There is a problem with the video, I can not watch it!!

Could you please upload it again, I really need some help on this video topic!!

Thanks

[Gaskination]

@AtyDeh

It definitely works. I'm sorry you are not able to watch it. There may be filter issues at your location. If you are trying to access it from work, then they probably block youtube. If you cannot access it from home, then you probably just need to try again another time. In the meantime, feel free to refer to my wiki for info on model fit: statwiki. kolobkreations. com

[chaitanya183]

found them all. That is fantastic!! Thank You :)

[chaitanya183]

Yes, I saw your statwiki as well. I am sharing it with all my fellow doctoral students. I already sent your videos to all. Can you also please share your data sets that you use for demos?

[Gaskination]

@chaitanya183

I've just received permission from the owner of the datasets. I'll post them on the wiki right now. They should be available within the next 20 minutes.

Enjoy!

[chaitanya183]

@Gaskination Thank you so much. I am going to share your tutorials with students at Information Systems department at Georgia State University. Really appreciate the work Sir.

[Gaskination]

@chaitanya183

Done. See the homepage of the wiki.

[chaitanya183]

If I could tell you how much I appreciate your work. Thank you so so much

[Gaskination]

@chaitanya183

Thanks! I'm glad it is helping someone. I plan on continuing to make more videos. I hope you have found my wiki as well: statwiki. kolobkreations. com

James

[juandv82]

Great video!! I'm trying to learn CFA on my own. I'm starting with validating a scale, as in your example. Is there any book or manual for beginners you recommend? Thanks!

[Gaskination]

@juandv82

The best book for concepts, measures, and thresholds is probably Hair et al 2010 Multivariate Data Analysis. There really isn't any good book for teaching the mechanics of performing statistical analyses, which is why I made my wiki: statwiki. kolobkreations. com

[juanchitus02]

@Gaskination thank you!!!

[juandv82]

@Gaskination Thanks!!!

[ecmlau]

Hi, your post is extremely useful. I have several questions. (1) I am sure there is no missing data in my analysis but I still cannot run GFI and RMR indices. Can you let me know how I can uncheck estimate means and intercepts. (2) I am doing path analysis but Amos does not allow me to check Modification indices? (3) Can you let me know how to do data screening please? Many thanks.

[Gaskination]

@ecmlau

click on VIEW, then ANALYSIS PROPERTIES, then in the window that pops up, click on the ESTIMATION tab. On the right side of the window there is a check box for ESTIMATE MEANS AND INTERCEPTS. This will also allow you to check modification indices (in the OUTPUT tab of the same window). For datascreening, see my wiki: statwiki. kolobkreations. com

[ecmlau]

@Gaskination

Hi, thank you so much. I can run GFI indices now after data cleaning. I have to go to the next step of reducing the model. Also need to interpret the numbers on the graph. Have you got links showing how I can do both? Thanks for your help.

[Gaskination]

@ecmlau

I don't understand what you mean by reducing the model, unless you mean to trim off the items that are not correlating very well with others. I don't have links for that, other than this video. But the general rule is that you want the average standardized loadings from items within a latent factor to be higher than the correlations between factors.

[ecmlau]

@Gaskination

Hi, many thanks. After triming, the resulting models generally have GFI

around 0.8+ and CFI 0.6 only . The highest is only GFI 0.845 but it never go further up. Every time I trim the model, both GFI and AGFI increase slightly but NFI, CFI, RFI, TFI went down at the same time. I doubt whether SEM model is the right tool for testing my model. Thanks for helping me.

[Gaskination]

@ecmlau

I am not familiar with your data, but the reason you may be observing poor fit might be due to misspecification of the constructs. Your constructs might be formative, rather than reflective. And if this is the case, then you may need to use partial least squares, rather than covariance based methods (like AMOS).

[ecmlau]

@Gaskination

I see formative in the sense that indicators are explanatory from indicators to constructs. My model is about the effect of personal values to behavioral intention of participation in continuing education through attitudes and subjective norm (Theory of Reasoned Action refers). I think this may be the problem with me. Is PLS similar to multiple regression or MANOVA? Actually I have little knowledge about statistical analysis and I followed your video to do the SEM. Many thanks.

[Gaskination]

@ecmlau

I highly recommend my wiki on PLS. It is just another form of SEM. It can handle formative constructs (unlike AMOS). My wiki is: statwiki. kolobkreation. com

It has several youtube videos linked to it to walk you through how to do an analysis.

[ecmlau]

@Gaskination  Hi James, I have stopped standardizing the regression lines with 1, then I can get a model with good model fit figures now. Thank you so much. I shall try PLS later on. Cheers!

[khajimarka]

My model is working now ...thaks to your video

[Gaskination]

what version of AMOS are you using? RMR and GFI should come out in the model fit section. The estimate means and intercepts cannot run at the same time as certain other options (like modification indices). So you need to uncheck it, but you can only do that if you aren't missing any data :)

[natala333]

Hi , this is extremely helpful - thank you! but i have 1 problem, my amos doesn t report RMR and GFI table in output. i know there is some problem with check estimate means and intercepts, but really do not know how to uncheck it and then calculate estimates, because there is error. could you maybe please help me? thanx!