innsbruck!

Thanks for visiting Kathy! I'm glad they are helpful. By the way, don't forget to visit my most recent videos on CFA using AMOS: CFA video (2020): kzfaq.info/get/bejne/mMidgrCZls6ooHU.html Second-order CFA: kzfaq.info/get/bejne/qJqbm7qbzsu0cnk.html Best wishes!

Glad you found this useful! best wishes, Mike

You have to change the figure from an oval to a rectangle since that's how the software distinguish between observed from latent variables.

Hi Catherine, the trick is that the overall fit for your full SEM (i.e, path analysis with latent variable) model is not only impacted by the structural relations you specify among your latent variables, but also by the level of fit of your specified measurement model. I think you could adopt either approach you noted when reporting on your results. If the item is causing serious problems with fit in your measurement model, then you could delete it (as it would also create problems with fit of your full SEM model) and note why you did this. If it is not really causing serious problems you could theoretically leave it in. Either way, you'd want to indicate to the reader that the item is not functioning as intended in measuring your construct. Hope this helps.

Hi Mike, thanks so much for your reply, I think I will delete the uncooperative variable. Your help is really appreciated :)

Technically there is no "data" for a latent variable, as it is just defined by the measurement model (as is being demonstrated in the video). Once you fit the measurement model, you can use SEM (structural equation modeling) to draw paths to other latent variables.

Hi there. If you want to specify orthogonal factors, just leave out the double-headed arrows connecting the factors. Of course, this will make your test of orthogonal factors pretty stringent because you are treating the correlation as zero.

Hi there. CFA is a particular type of structural equation modeling. When you say "no relationships like a structural equation model" I'm assuming you are referring to a path analysis using latent variables. Models using CFA are essentially measurement models, where the focus is on testing a proposed factor structure involving a set of manifest variables. In effect, you are addressing the question, "What factor model explains the association among a set of observed variables?" Unlike exploratory factor analysis, where the focus is on building a factor model, CFA is aimed at testing a proposed factor structure. Oftentimes, CFA is carried out en route to testing a path model involving latent variables. It is generally pointless to carry out path analysis with latent variables if your measurement model (assessed via CFA) exhibits low fit. If your measurement model fits the data well, then it makes sense to move towards specifying predictive relationships among your latent variables. Hope this helps!

If the p-value is not significant, then it indicates that the item may not be a good indicator of the latent construct. I don't know if there is any one thing "to do" when this occurs. If you are trying to maximize fit to the sample data, you could test alternative specifications in an exploratory fashion (e.g., maybe modification indices?) to see if the item might load better on a different factor. Or you could remove the item if it is generally problematic for the model and see if fit improves. The only thing about these kinds of empirical strategies is that you are likely going to end up overfitting your model to the data (reducing generalizability of results to other population samples). Really, the question you should be considering is "Does the item I have written adequately reflect the construct I am trying to measure?" If a coefficient is non-significant, then it would appear you have evidence that this is not the case. When this occurs, you may need to either rewrite the question if it does not adequately reflect the construct definition and/or eliminate it if it is unrelated to the definition and/or include new items that do a better job reflecting the construct definition. I could get into a long protracted conversation regarding construct validity, but these are just a few thoughts I had to keep things manageable in my reply. Hope this helps. Best wishes.

@@mikecrowson2462 Thank you for your explanation. Actually, I try to validate a scale into Indonesian language. I look through the data again and try your suggestion. Thanks

Hi there, the NNFI is called Tucker-Lewis Index (TLI) in AMOS. They are the same.

The data set is provided as a link under the video description.

Hi there. Funny you should ask. I just uploaded a new version of CFA with Amos that includes a powerpoint you can download (following the link underneath the video description). There are citations in it. You can visit the video and get the Powerpoint here: kzfaq.info/get/bejne/mMidgrCZls6ooHU.html Thanks for your comment!

By the way, the CFI & TLI at .95 are generally regarded as more consistent with closer fit. However, in the powerpoint I include references, where .90 or above may be considered acceptable.

It's ok I fixed it now by removing a rogue item and replacing it with another one that also correlated

You're most welcome! Also, just so you know, I do have a much more recent video on CFA with AMOS here (kzfaq.info/get/bejne/mMidgrCZls6ooHU.html). It contains more updated information and supplemental information.