It is a model where only the total scores of each variable are used.

As I know, for any analysis (direct or mediation) using AMOS, items need to be loaded onto construct. In fact, this is the very basics of SEM. In some online tutorials on Youtube, sometimes items are not displayed for the sake of simplicity in presentation and understanding. This is what I think, if I correctly understand your question.

All the best.

Hello Shayan,

honestly, I did not really understand your question? Could you try again and be more explicit? What is your model (IV, Mediator and DV). How do you measure these?

Best,

Holger

Thank you so much for your answers. I am very grateful.

I have 2 IVs, 3 mediators and 1 DV.

Between 4-10 items are loading on each construct.

This successfully allows me to have the results for a measurement model and a structural model.

However, when I need to follow this up with mediation analysis (bootstrapping), in the tutorial videos I always see the analysis between the constructs without items loading onto each construct. What I am not clear about is whether I should calculate the means of items to have an aggregate figure for each construct or what I should do to have the items loading onto each construct. For example, in the tutorial videos below, items are not loading onto each construct for analysis. My question is how to conduct mediation analysis without each item loading on the construct. Would for example, total scores be used or means or medians.

These are two examples of the tutorials:

https://youtu.be/j_yufPUjkwk

https://youtu.be/eWHqxrsI0NI

Many thanks for your guidance and answers!

Ah, I see. Again the videos by Gaskin ;)

In the video, he conducts a path analysis with observed variables. This can be single items or composites (from sums or means of several items, which does not matter).

I am no AMOS user. Try to follow his steps to bootstrapp the standard errors of the indirect effect. I would assume they work for latent variables with several indicators, too.

Better would be to skip AMOS altogether and switch to R / lavaan (lavaan.org). I know this seems like a huge step but after a few days you will be able to do so much more. And I would offer help (as I am responsible then :D)

Beyond that, I would be interested how your model exactly look like, especially what you did to reflect the correlations among mediators. How was the model chi square test? How was the question wording of the items; how was the structure. I know I am very curious. Choose to present anything depending on phase of problems you are faced with.

Perhaps this paper is helpful:

VanderWeele, T., & Vansteelandt, S. (2014). Mediation analysis with multiple mediators. Epidemiologic methods, 2(1), 95-115.

With regard to modeling the constructs: Please avoid the error to press each and every set of items in to a common factor model. Sometime sets of items rather represent an aggregate construct. Then testing a common factor model is wrong and will lead to bad fit. Then, adding all together may be a better choice (although it has some problems). And no, SEM does not require factors, not even latent variables. You are very flexible.

And finally, some cautious words. If your model looks like the model in Gaskins video (a partial mediation model), you can unfortunately not really test the mediation model, and more importantly, a structure has tons of alternative models that describe the data as your model does. That is, you can not really say much about mediation. For instance, I simple tripartite model with X,M,Y has over 40(!) alternative models.

Having a full mediation model, in contrast, implies only a handful. That is, if you manage it to have the effects of your IVs on the DV fixed to zero (and this restriction holds by the chi square test) than much is gained. If you don't, you would need further variables to identify the effects (called instrumental variables). Beyond that, you need appropriate control variables to reduce the danger of confounding bias. Both for the X->M link and for the M --> Y link.

I know, these are no good news, but the status of mediation analysis in the social sciences is the same desaster as it is for testing factor models.

In this thread, I wrote something about assumptions of the mediation model:

https://www.researchgate.net/post/I_want_to_know_the_assumptions_of_multiple_mediation_analysis_with_AMOS_output_Can_someone_give_any_guidelines

Best,

Holger

Dear Holger,

Thank you so much for such a comprehensive response!

The paper and the thread you shared are really useful. I am going to read the paper in detail and look at the model again in context of all the information you have shared. I am hopeful I should be able to get some results.

About R, I have been thinking the same. If not right now, for the next project I am definitely going to learn R programming. Thank you so much for offering your help, you are too kind!

Kind regards,

Shayan

Hello Shayan Shaikh , with reference to your question, I can relate my personal experience of the confusion faced by most new users (researchers) of statistical analysis research tools (e.g., AMOS), especially after going through initial learning efforts on YouTube, regarding mediation, moderation, CFA (and EFA); where all of the focus remains on statistical analysis output only i.e., full mediation or partial mediation etc. Most of the times the theoretical foundation for the same is ignored.

Anyways, as you have decided to stick with AMOS for your current project and would opt for R later, as suggested by respected Dr. Holger Steinmetz , according to my limited knowledge I would recommend few very useful videos by Dr. Mike Crowson, which are I think exact match for your concern.

You can get Factor Scores (sometimes called Latent Variable Scores) to use in structural model analysis, in the same way you have seen in Gaskin's observed variable videos, while doing CFA or measurement model analysis.

https://youtu.be/8DtZInumbmQ

You can perform measurement as well as structural model analysis with latent variables by keeping the construct items in the model.

https://youtu.be/p1pGLV7FIaI

Another useful like for mediation model analysis in AMOS: https://youtu.be/MQR0kLXDqhk

Hope it will help.

Regards

Dear Muhammad,

Thank you so much for your comprehensive response. This will really help.

Warm regards,