Hi James, thank for all your videos, Always helping me. Let me ask two things: In a multigroup analyses using SEM It´s good to report a correlation table between factors? Another thing that I noticed in articles is that correlation tables are reported when multigroup analyses were done with CFA instead of a more complex model (e.g., SEM). In case of SEM, there is any other table to be reported in place of correlations table in a multigroup analyses? Thank you very much and congrats for all your excelent work, Cesar
It is always good practice to report the correlation table. When conducting multigroup analysis, it is best to report that correlation table for each group. Because the structural model does not have all factors/variable covaried, the correlation table is most often reported in association with the measurement model (CFA).
Hi James Thank you so much for everything that you post on youtube. This is making my entire PhD life so much easier! :) I just have one question re the video Serial Mediation in AMOS. How would I add a moderator to this kind of model? And then also measure the indirect effects etc? So for example, if I had a moderator on path B, how can I also include and measure that? Any help you can offer would be greatly appreciated.
If the moderator is an interaction, and it is with an endogenous variable, then you would have to covary the moderator and interaction with the error of the other variable. If instead it is a grouping moderator, then it is super easy. Just assess the difference in estimates for path B across the two groups.
Hello, Dr. Gaskin. When we have several exogenous variables (not just topical interest), should we do the serial mediation analysis one by one for each (e.g., deleting path A from Topical Interest and inserting it for another exogenous variable)?
Dear James, I need to ask if we have multiple independent variables then will we perform serial mediation independently for every variable by denoting paths or together by drawing the covariance between IVs??? Please advise
Serial mediation is only applicable if your variables have effects in a particular order. However, if all of your IVs are covariates, rather than dependent upon each other, then just covary them.
Professor, can u explain how to interpret a counter intuitive -ve beta coefficient in mediation. In my theory training->ethics->commitment, my I.V-D.V, M.V-D.V beta values turned out be negative. So in my defence if I have to interpret this , is there way other than saying that " I ignored a potenial moderating variable which would have explained the -ve beta.?
Since mediation is just one path multiplied by another path, the direction (positive/negative) of the mediation is theoretically tricky because if one path in the mediation is negative, then the resulting product will be negative. So, I would read more into the size rather than the direction of the mediation. If you observe a negative indirect effect and a positive direct effect, then this can be considered "competitive" mediation: Zhao, X., Lynch, J. G., & Chen, Q. (2010). Reconsidering Baron and Kenny: Myths and truths about mediation analysis. Journal of consumer research, 37(2), 197-206.
Hi, a 03 item variable (intentions) has AVE (0.4) < MSV (0.508). I have tried item deletion but didn't work. CR 0.666, Cronbach (0.672), factor loadings >0.6 for all items, HTMT is good and model fit for measurement and structural model was easily achieved. I can't eliminate this variable. Pls suggest a remedy, thanks.
Dear Dr. Gaskin, greetiings, my model is with 2 mediators so I have Path A, B, C. Even without using estimands, the model calculates the same coefficients on these paths. So what exactly is the benefit of using estimand, is that only to assess the significance of the path ?
The benefit of the estimand is that you can specify sub-paths to evaluate. For example, in the video above, what if I wanted to check the full serial mediation versus the mediation without going through exploration and discovery? The estimand lets me specify that, whereas the default features in AMOS will only estimate the full indirect path.
Dear Professor Gaskin, can semantic differential questions be used for SEM analysis? In addition, is it possible to use semantic differential and Likert scale questions in a questionnaire when wanting to use SEM analysis?
Yes and yes. However, you just have to be careful in the interpretation. Treat a positive effect as an increase in the right side of the scale, and a negative effect as an increase in the left side of the scale. Or, if the two sides of the scale are truly opposite, then just name the construct based on the right side of the scale and you can interpret estimates the same way you would for Likert scales.
Hello Sir ! Thanks For sharing a very useful tool to run serial mediation. just one little thing that it give us unstandardized Beta. how we can have the standardized beta. Thanks and Best regards
currently the only way to do that would be to multiply the standardized indirect paths. Or, if you standardize all the variables before running the analysis, then the unstandardized estimate will be the same (or very close) to the standardized estimate.
Dear Dr; Gaskin, I can't seem to find the plugin for serial mediation (AMOS21). I went to statwiki with no luck. Would it be possible for you to post a link to download the file. Thank you in advance and thank you for your work!
Hello sir. I have one small query. Should i run serial mediation using imputed values or by using my latent constructs with indicators? Which one is better? Thank you sir.
In general, latent factors retain more information, and therefore they are considered more precise. It is common practice, for the sake of reducing complexity, to use factor scores as adequate proxies for latent factors. However, you may notice differences in outcomes.
@@Gaskination Thank you sir for the clarity. Sir, just one more doubt. Can we calculate standardised indirect effects instead of unstandardised indirect effects using user estimands?
DF are the number of parameters (paths) that could have been estimated, but were not included. So, if you have only one dependent variable and no mediators, you will always have zero DF (if variables are observed/manifest rather than latent). If you have more than one DV or if you also have mediators, then if there are any paths not included in the model, these will increase your DF. If all possible paths are included, you will have zero DF.
Hello sir. I have been using your plugin. But sir the plugin does not work with second order constructs and only works with imputed value and first order constructs. Kindly suggest
@@Gaskination sir what I did was I calculated imputed values for first orders and then I took the imputed values as latent factor for second order constructs. Accordingly my model became a first order model. So then when I ran the test I got the results. Although there is a very slight difference in the beta values that I got from Amos and from the estimand.
Thank you Gaskin. I have always use your lectures for my research work. Thanks again,
Good work James!!! Thank you so much
Thank you so much for your effort.
Thank you sir..
Thank you
Thank you so much for the estimand syntax. In serial mediation, do we still need to check for CFI and CMIN/df?
Thank you, Dr. Gaskin. This is great!
Does it work with a latent model too?
Yes. This should work just fine with a latent model.
Thank you, Gaskin. It's a useful video!! Does it work on a second-order model too?
Yes, the estimand will work for any causal model with indirect effects.
Hi James, thank for all your videos, Always helping me. Let me ask two things: In a multigroup analyses using SEM It´s good to report a correlation table between factors? Another thing that I noticed in articles is that correlation tables are reported when multigroup analyses were done with CFA instead of a more complex model (e.g., SEM). In case of SEM, there is any other table to be reported in place of correlations table in a multigroup analyses? Thank you very much and congrats for all your excelent work, Cesar
It is always good practice to report the correlation table. When conducting multigroup analysis, it is best to report that correlation table for each group. Because the structural model does not have all factors/variable covaried, the correlation table is most often reported in association with the measurement model (CFA).
@@Gaskination Thank you very much my friend!
Thank you Professor. I sent an email with two questions. Hope you can kindly answer it. Thank you
Hi James
Thank you so much for everything that you post on youtube. This is making my entire PhD life so much easier! :)
I just have one question re the video Serial Mediation in AMOS. How would I add a moderator to this kind of model? And then also measure the indirect effects etc? So for example, if I had a moderator on path B, how can I also include and measure that?
Any help you can offer would be greatly appreciated.
If the moderator is an interaction, and it is with an endogenous variable, then you would have to covary the moderator and interaction with the error of the other variable. If instead it is a grouping moderator, then it is super easy. Just assess the difference in estimates for path B across the two groups.
@@Gaskination Thank you so much!
Hello, Dr. Gaskin. When we have several exogenous variables (not just topical interest), should we do the serial mediation analysis one by one for each (e.g., deleting path A from Topical Interest and inserting it for another exogenous variable)?
It is best to include all the variables. We are likely to increase false positives if we ignore paths or variables.
@@Gaskination thank you.
Dear James, I need to ask if we have multiple independent variables then will we perform serial mediation independently for every variable by denoting paths or together by drawing the covariance between IVs??? Please advise
Serial mediation is only applicable if your variables have effects in a particular order. However, if all of your IVs are covariates, rather than dependent upon each other, then just covary them.
Professor, can u explain how to interpret a counter intuitive -ve beta coefficient in mediation. In my theory training->ethics->commitment, my I.V-D.V, M.V-D.V beta values turned out be negative. So in my defence if I have to interpret this , is there way other than saying that " I ignored a potenial moderating variable which would have explained the -ve beta.?
Since mediation is just one path multiplied by another path, the direction (positive/negative) of the mediation is theoretically tricky because if one path in the mediation is negative, then the resulting product will be negative. So, I would read more into the size rather than the direction of the mediation. If you observe a negative indirect effect and a positive direct effect, then this can be considered "competitive" mediation: Zhao, X., Lynch, J. G., & Chen, Q. (2010). Reconsidering Baron and Kenny: Myths and truths about mediation analysis. Journal of consumer research, 37(2), 197-206.
Great mediation video, I feel so relaxed that I'm going to sell all my weed, thank you, so much, let's be RUclips pals?
I'm not sure why this comment keeps popping up to the top. I'm not sure how to be youtube pals. You are welcome to subscribe.
Hi, a 03 item variable (intentions) has AVE (0.4) < MSV (0.508). I have tried item deletion but didn't work. CR 0.666, Cronbach (0.672), factor loadings >0.6 for all items, HTMT is good and model fit for measurement and structural model was easily achieved. I can't eliminate this variable. Pls suggest a remedy, thanks.
Consider whether this factor is actually measured formatively.
Dear Dr. Gaskin, greetiings, my model is with 2 mediators so I have Path A, B, C. Even without using estimands, the model calculates the same coefficients on these paths. So what exactly is the benefit of using estimand, is that only to assess the significance of the path ?
The benefit of the estimand is that you can specify sub-paths to evaluate. For example, in the video above, what if I wanted to check the full serial mediation versus the mediation without going through exploration and discovery? The estimand lets me specify that, whereas the default features in AMOS will only estimate the full indirect path.
Hi James,
The Model Fit Measures (ModelFit.dll) is not work with AMOS 26. Did you test it?
Yikes! You're right. It doesn't work for me either... I'll have my RA look at it.
Thank you Prof James. I have emailed you a question. I'm appreciate if you would like to answer it. Thanks
Hey, can you please distribute his email ?
I need it so bad
Hey can u please distribute his email address ?
I need it so bad
@@hajerjmal9525 james.eric.gaskin@gmail.com
Dear Professor Gaskin, can semantic differential questions be used for SEM analysis? In addition, is it possible to use semantic differential and Likert scale questions in a questionnaire when wanting to use SEM analysis?
Yes and yes. However, you just have to be careful in the interpretation. Treat a positive effect as an increase in the right side of the scale, and a negative effect as an increase in the left side of the scale. Or, if the two sides of the scale are truly opposite, then just name the construct based on the right side of the scale and you can interpret estimates the same way you would for Likert scales.
Brilliant. Professor@@Gaskination, thank you so so much for your kind assistance and cooperation.
Thank you so much DR. Gaskin. How I can download estimating serial mediation?
check the statwiki: statwiki.gaskination.com/index.php?title=Plugins#GENERAL_INSTRUCTIONS:
Hello Sir ! Thanks For sharing a very useful tool to run serial mediation. just one little thing that it give us unstandardized Beta. how we can have the standardized beta. Thanks and Best regards
currently the only way to do that would be to multiply the standardized indirect paths. Or, if you standardize all the variables before running the analysis, then the unstandardized estimate will be the same (or very close) to the standardized estimate.
@@Gaskination Thank You very much sir :) stay blessed :)
Dear Dr; Gaskin, I can't seem to find the plugin for serial mediation (AMOS21). I went to statwiki with no luck. Would it be possible for you to post a link to download the file. Thank you in advance and thank you for your work!
There is no plugin for serial mediation. There is only the estimand.
Thank you Professor for your videos. I am not able to edit the estimand. A pop up window shows an error. Can you please help me with this. Thanks
Make sure to put it into an editable folder, like the downloads folder or documents folder, but not in the plugins folder.
@@Gaskination Many thanks Sir
Dear James Gaskin, is there any paper which has used this serial model mediation please?, with thanks in advance
Probably. But none of mine yet... Here are some useful search results: scholar.google.com/scholar?hl=en&as_sdt=0%2C45&q=%22serial+mediation%22&btnG=
Hello sir. I have one small query. Should i run serial mediation using imputed values or by using my latent constructs with indicators? Which one is better? Thank you sir.
In general, latent factors retain more information, and therefore they are considered more precise. It is common practice, for the sake of reducing complexity, to use factor scores as adequate proxies for latent factors. However, you may notice differences in outcomes.
@@Gaskination Thank you sir for the clarity.
Sir, just one more doubt. Can we calculate standardised indirect effects instead of unstandardised indirect effects using user estimands?
@@dr.anupreetkaurmokha5792 Yes, with this other plugin: ruclips.net/video/-xzkjqKhWHE/видео.html
@@Gaskination Thank you sir
why every time i run the data in AMOS, the DF always zero?
DF are the number of parameters (paths) that could have been estimated, but were not included. So, if you have only one dependent variable and no mediators, you will always have zero DF (if variables are observed/manifest rather than latent). If you have more than one DV or if you also have mediators, then if there are any paths not included in the model, these will increase your DF. If all possible paths are included, you will have zero DF.
Hello sir. I have been using your plugin. But sir the plugin does not work with second order constructs and only works with imputed value and first order constructs. Kindly suggest
Correct. For those other types of models, you'll need to use just the estimand shown in the video above.
@@Gaskination sir what I did was I calculated imputed values for first orders and then I took the imputed values as latent factor for second order constructs. Accordingly my model became a first order model. So then when I ran the test I got the results. Although there is a very slight difference in the beta values that I got from Amos and from the estimand.
@@fodieebydefault That will work as well.
@@Gaskination thank you sir
Can you explain me how to run serial mediation with second stage
moderator in AMOS?
Just do as normal, but covary the interaction with the error of the mediator it interacts with since you can't covary it with the mediator.
@@Gaskination thank you so much