How to Test Multiple Mediators using AMOS (Structural Equation Modeling)

[@larissacury7714]

Amazing!! Thank you very much!

[@somebodyyouusedtoknow11]

Hi Prof. Collier, for some reason the text tab in Object properties is empty. I cannot label the path.

[@premiereducare]

Professor, this is a very great explanation. How can we get this Customer Delight Data. Sav file ? Professor, I got your book. Please help me to find the dataset to practice myself.

[@joelcollier9387]

Give me an email address and I'll send it to you. Cheers.

[@DokoloCity]

Hi Prof Collier, thanks, I bought the book... is there a chance you can run a tutorial for all the chapters in the book via Amos..

[@joelcollier9387]

Thanks for supporting the book. I am in talks to run some 2 day workshops in November. The KZbin tutorials for a whole chapter might not coming until after that. I will post smaller/quicker videos in the meantime.

[@muhammadaamirkhan8182]

Though, I followed your mentioned procedure. However, I'm getting an error while testing/saving the estimand. The error is "Keyword is not valid as indentifier. 'Mod' is not a member of 'v'. Keyword is not valid as identifier". Please advise which keyword(s) should be worked on to remove the error, thanks.

[@joelcollier9387]

If you are getting that error it is most likely because the names of your constructs in the graphics window do not mirror the names in the estimand function. I would check those. You might also want to check that you do not have spaces between the words in your estimand descriptor line. If you need to use a space, you might want to use the underline character instead.

[@eylemtasdagitici]

Dear Prof. Collier, If the independent variable is a scale consisting of 4 dimensions with 5 statements in each dimension, should we take the average of 20 items and analyze them? Or should we take the average of each dimension separately and evaluate and analyze 5 dimensions as 5 statements? Or should we include 20 expressions in the analysis as one dimension? I am in serious trouble at this point. Your answer is very valuable to me. Thank you.

[@joelcollier9387]

So, the best and most appropriate way to handle this is a higher (second) order model where you model all the constructs and items individually. I have another video on higher order models if you need assistance. kzbin.info/www/bejne/aXiXYaSmiclji8k If you are going to use summated constructs, then you can only summate items within a construct. Do not summate numerous constructs into one overarching concept. That is conceptually inappropriate to do.

[@fatihcelik7537]

Hello dear Collier. For the mediating effect in Multiple Mediators analysis, S.E. and C.R. Can we report (t) values? These values are not available in your indirect impact analysis. Also, can we standardize the regression coefficient for the mediation effect?

[@joelcollier9387]

when running a bootstrap analysis, you will not get a t-value because it is resampling your data over and over. That is why it gives you a upper and lower bound confidence interval. If your confidence interval does not include zero, you have a significant indirect effect (p-values will be listed). So long story short...it will not provide you a t-value...and you don't really need it either. As for standardization, when you are running a bootstrap analysis, standardizing the data is not ideal way to present your results. You want to present the unstandardized estimates especially from a bootstrap perspective.

[@fatihcelik7537]

@@joelcollier9387 As a result of the Bootstrap test, when we use unstandardized coefficients in the structural model, do you recommend using unstandardized factor loads when we test the measurement model (CFA)?

[@joelcollier9387]

@@fatihcelik7537 No a separate CFA should be reported with standardized regression weights to allow for comparison

[@gemini1404]

Had a question about mediation analysis if you don't mind.. I have a one tailed hypothesis and want to set my alpha value at 0.025. Do I set the confidence interval at 97.5 in Amos? Can you please guide me as to how to go ahead with this since AMOS gives two tailed significance value

[@joelcollier9387]

So if you are only concerned with one direction, a confidence interval set to 90% would give you a one tale test at .05. If you are wanting a .025 significance, then you would leave the bias corrected confidence interval set to 95%. (as shown in the mediation video above). Again, this is only appropriate if you have a one tail test and are willing to ignore the other side of the distribution. This can be set in the analysis properties window under the "bootstrap" tab. Hope this helps.

[@gemini1404]

@@joelcollier9387 thank you. But since my hypothesis is one tailed I'm confused about how would I interpret the two tailed significance value given? Do I half the two tailed significance value and compare it to 0.025?

[@gemini1404]

@@joelcollier9387 this is something that I'm confused about since AMOS does not provide one tailed significance

[@joelcollier9387]

@@gemini1404 You are correct that AMOS will not give you a one tail significance test. To find the exact significance, you will take the two tail test and then half that value (you are only concerned with one side of the distribution). So if you found a value of .05 with a two tail test that is the equivalent of a .025 value for a one tail test.

[@gemini1404]

@@joelcollier9387 I really appreciate you responding to all my queries. I am a novice in statistics so all of yours answers and videos help alot. I have just one last question if its okay with you. My hypothesis is essentially that "Attitude, norms and control predict intentions to engage in a behavior." Would I really even require a one tailed p value to test this hypothesis? I observed in some cases that people have interpreted similar hypotheses using bootstrapping and a two tailed p value.

[@florisvandervelde8819]

Hi sir, what is the benefit of using 2 mediators at the same time, compared to analyzing them separately?

[@joelcollier9387]

You see the collective effect on your ultimate dependent variables instead of piecemealing the individual mediators influence. It just gives you a better understanding of how the dependent variables are influenced from a comprehensive perspective.

[@fadhlurrahimazmi8257]

I have multiple DV and multiple mediators. Can I use this technique as well?

[@joelcollier9387]

Absolutely. Just label all the paths and use the estimands to test the ones you are concerned with

[@fadhlurrahimazmi8257]

@@joelcollier9387 Thank you Prof.

[@ladydija3228]

Please can we test multiple mediators in STATA ? My model contains 4 mediators is that possible ?

[@joelcollier9387]

I can't speak to STATA . I have never used it but as for AMOS you most certainly can use 4 mediators in a model.

[@ladydija3228]

@@joelcollier9387 thank you for your response

[@andracoman3289]

Can I test multiple mediation with amos 23 or just with amos 24?

[@joelcollier9387]

You can test it with AMOS 23 as well. The Estimands function is included in AMOS 23 and works the exact same way as version 24.

[@andracoman3289]

@@joelcollier9387 thanks for your answer!

[@商業智慧學院博士班羅]

indirecttest=A*B ....._indirecttest=X*Y