SmartPLS 3 Mediation

[@Gaskination]

Here's a fun pet project I've been working on: udreamed.com/. It is a dream analytics app. Here is the KZbin channel where we post a new video almost three times per week: kzbin.info/door/iujxblFduQz8V4xHjMzyzQ Also available on iOS: apps.apple.com/us/app/udreamed/id1054428074 And Android: play.google.com/store/apps/details?id=com.unconsciouscognitioninc.unconsciouscognition&hl=en Check it out! Thanks!

[@LockyLawPhD]

Thanks to your suggestion to SmartPLS 3, Specific Indirect Effects have been added to the software.

[@YenkuKuo]

Dear Dr. Gaskin, could you tell me what recorder software you use in this video? Thanks!

[@Gaskination]

I use screencast-o-matic. I love it. It has a built-in video editor.

[@bushrarao1869]

Very informative and easy to understand👍

[@ishpreetvirdi4040]

HELLO DR. i am working on mediation, my model is second order construct (reflective- formative). on testing mediation, my IV is that second order construct. so, should i use the latent variable scores for testing mediation also (calculated earlier in my full model testing) or should i use the first order variable indicators/ values (for that particular consrtuct which is second order) for testing mediation? Actually, i tried both ways, with the help of your video, and i am getting partial mediation (as calculated using latent variable scores) and full mediation otherwise. Which method should i follow? This is the last phase of my research work. plz suggest.

[@Gaskination]

If the 2nd order factor isn't endogenous, then you can use the fully latent model (not the latent variable scores). The only time you need to use latent variable scores is when a 2nd order formative factor is endogenous (like as a mediator or DV).

[@yensin2933]

Dear Prof., my model having 3 IV, 1 mediator and 1 DV, I would like to use the variance accounted for (VAF) method, is it correct that I have to rerun the model separately with IV one by one to get the indirect effect and direct effect?

[@Gaskination]

I'm not familiar with the VAF method. I'd recommend following whatever the literature says.

[@helenaliao5757]

Dear Prof. Gaskin, I saw some papers using Confidence interval to check the significance of the indirect effect (not include 0), and I found the significance on P value vs Confidence interval (bias corrected) is not consistent in my analysis. Which one will you suggest me to follow? And does the "sample mean" in the path estimate stand for "beta"? What is the number stand for "point estimate" for the indirect effect? Thanks a lot for the tutorial and the reply.

[@Gaskination]

1. If the confidence interval includes zero, then the p-value should not be significant. However, the size of the confidence interval matters. If you are using 95% confidence, then the interval will span larger than if you are using 90% confidence. By default, SmartPLS 3 uses the 95% confidence interval, so if the interval does not contain zero, then it means it is significant at the alpha=0.05 level. 2. Sample mean is the mean of betas across all the bootstrap resamples. 3. I'm not sure about point estimate.

[@helenaliao5757]

Dr Gaskin, Thank you for replying me such a basic question, prof..

[@eimanal-khalaf9377]

Thanks for the tutorial ! I have question , in my structural model I have figured out that one IV is directly significant to the target DV and has no mediation , so what shall I do with the paths from the IV to the mediator after obtaining the results so I have to keep it or deleted it and carry out bootstraping procedures again? Thank you

[@Gaskination]

I recommend keeping it, if it is part of your theory.

[@elizabethcarolina2473]

Dear Mr. Gaskin, Thank you for your tutorial. What's the difference between PLS Algorithm and Consistent PLS Algorithm? When should I use them for my mediations model? Thank you.

[@Gaskination]

Use PLS algorithm for all analyses unless you have all reflective factors. PLSc is simply to replicate (be consistent with) covariance based SEM algorithms.

[@RoamingEveryday]

Thanks for sharing this. Seems like the mediation analysis you have presented is according to Barren and Kenny's recommendations. However, is it really important to decide whether there is a partial or full mediation by considering the significance of direct relationships (in the presence of mediator) while following Preacher and Hayes? I'm asking this because I have read that the decision in P&H mediation analysis is not based on the significance of direct relationships (IVs & DV) but on the Value accounted For (VAF).

[@Gaskination]

I agree. I ran this analysis using P&H initially, but then added the comments about "partial" and "full", as some journals still expect it.

[@poornisakrabani1563]

Hi Dr Gaskins, when I clicked on specific effects, I got this A -B-C-D. This is the first time that I have seen two mediators. How do I explain this? Thanks

[@Gaskination]

The specific indirect effects shows all indirect paths, including serial paths.

[@poornisakrabani1563]

Thank you. I just found out that it's called serial mediation. Could you please do a video on that? I find your videos short and very easy to understand.

[@Gaskination]

@@poornisakrabani1563 Serial mediation is just the multiplication of all subpaths for an indirect path. I have an AMOS plugin for it: kzbin.info/www/bejne/Y6ndnJ2ngM2Kfqc

[@poornisakrabani1563]

@@Gaskination Tqvm Dr Gaskin

[@shra5126]

Hi Dr Gaskin, How can I find if there is a competing or complementary mediating effect? Is there any way in SmartPLS? Do you have any video related to that?

[@Gaskination]

Check out: 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.

[@marleyalvory3044]

Hi good day to u Dr. Gaskin. I would like to know when I report the results, should I use original sample or sample mean? 😊

[@Gaskination]

When bootstrapping, it is customary to us the sample mean, as this is the more robust result (it is the average coefficient across all subsamples when bootstrapping).

[@marleyalvory3044]

@@Gaskination Thank you for your reply :)

[@researchmethods1322]

Dear Dr. Gaskin. May I know why the subsample is 1000? I got confuse when some said 500 and some suggest 5000. But if I have collected 400 data, what is the subsample I should put? Hope to hear from you soon :)

[@Gaskination]

It is just a trade-off between rigor and compute time. More subsamples is more rigorous but takes more time. You also get diminishing returns after 50 actually. 100 is often enough (according to the original creators of the bootstrap method). However, contemporary hardware enables far more subsamples. There are published recommendations of 1000, 2000, 5000, 10000, 19000 (not sure why on this one...), and more. I usually do 2000. If you run it again with more or fewer, and observe nearly identical results, then you know your sample and model are robust to subsamples of varying size.

[@researchmethods1322]

@@Gaskination oh I see thank u for explaining Dr. Gaskin. Wish you have a great day. 😄

[@chefberrypassionateresearcher]

Sir, for my indirect effect, my confidence intervals align with my p value's significance. But 'confidence interval bias corrected' is not matching with the p value significance. Can I report confidence intervals, instead of bias corrected confidence intervals?

[@Gaskination]

You can report either or both as long as you clarify which it is.

[@chefberrypassionateresearcher]

@@Gaskination Sir, what is the difference between confidence interval and "bias corrected confidence interval". How can I justify in mediation analysis, if I want to mention confidence interval, and not the "bias corrected confidence interval"

[@Gaskination]

@@chefberrypassionateresearcher Bias corrected simply adjusts the confidence intervals (usually dampens) to correct for artificial inflation. It is very subtle, but if your estimates are on the edge (as in your case), it can make a difference. You do not need to justify not using bias corrected intervals. You do not even need to mention bias correction unless a reviewer requests it.

[@ammaralrawahna5993]

Dr. Gaskin Hello I have model with 5 IVs and a mediation variable and DV . i am interested to assess the mediation relationship for one of IVs not all, with DV. so, when i examined the relationship of mediation for that IV without including other IVs, the result of mediation was significant. but when all the IVs included in the analysis the relationship of mediation for that IV turned to be insignificant.My question, What is the reason to switch the relationship ? sincerely

[@Gaskination]

Ammar Alrawahna it’s because regression accounts for shared explained variance. This is perfectly natural

[@maheenkhan630]

Dear James Gaskin ! i have a variable of academic achievement as a DV . how can I make indicators of academic achievement to run the analysis ?

[@Gaskination]

If you are asking how to measure academic achievement, I would recommend an objective approach using grades or GPA.

[@suha3890]

Thank you for doing the SmartPLS series...can i test mediating in SmartPls for different IVs and one mediating variable to the same DV one time? then i follow sam steps that you explained in your tutorial? or i have to spilt the model to IV---> MV---->DV then run bootstrapping and check indirect effect for each sub-model ?

[@Gaskination]

You can check them all at the same time. SmartPLS also recently emailed me to let me know they have implemented multiple mediation features into the next version of SmartPLS.

[@suha3890]

Thank you Dr for your valuable information May i ask you how can report mediating test in smartpls 3 in research paper ? any sample research paper?

[@suuuuuuuuuuus]

So you get the direct effects from IV to DV from the path coefficient table not from the total effects table? You said that if the direct effect in in significant and the indirect effects is significant this means that we have mediation. But the smart PLS book says that to assume mediation first the the direct effect path 13 from iv to dv should be significant if the mediator is not included in the model. My results show that my direct effects are not significant but my indirect effects is significant. I am worried after reading this from the book does that mean I don't have mediation?? Thanks, Sara

[@Gaskination]

Direct effects can come from the path coefficient table. If indirect is significant, then there is mediation. If you have a direct effect included, and it is significant as well as the indirect path, then it is partial mediation. If direct is non-significant, but indirect is significant, then it is full mediation (or "indirect only" mediation). See Zhao et al 2010 "reconsidering barron and kenny" for more details.

[@suuuuuuuuuuus]

James Gaskin thank you your kindness to replay. I have a serial mediation with 3 IVs and 2 DVs and 4 Mediators that are in a straight line- each one leads to the other (no parallel). The path coefficient (direct) is not significant. Then I have different numbers for the total effect, total indirect effect, and specific indirect effect. the total and total indirect numbers are not significant, but the specific indirect effects are significant! This means I still have mediation, right? I am confused when reporting the mediation results I have to put path c (mediated model) and path c' (non mediated model) but the problem is I don't know should I report the total effect or the indirect total effect or the specific indirect effects. They are not the same they have different values. Thank you for your cooperation.

[@Gaskination]

@@suuuuuuuuuuus Yes. If specific indirect effects are significant, then the mediation is present for the variables involved in that relationship. As for how to report mediation, check any well-known paper on mediation. Here are a few: statwiki.kolobkreations.com/index.php?title=References#Mediation

[@louism1971]

Dear Dr.Gaskin, I use smart pls 3.2.7 and if i found all the indirect effect and direct effect is significant can I interpret there is a mediation? even using baron and kenny approach?

[@Gaskination]

If direct and indirect effects are significant, then Baron and Kenny would call it "partial mediation".

[@louism1971]

James Gaskin thank you very much with your help I am able to finished my thesis examination.

[@kang-tingwang9245]

Dear Mr. Gaskin, Your video is quite useful and easily understandable! Additional query is about "moderated mediation" with SmartPLS 3. My reflective model is asked to exert moderated mediation. Would you suggest a series of steps about conducting such process? Thank you!

[@Gaskination]

I don't have a video for that yet. The easiest way is to test an indirect effect while assessing group differences (where the grouping variable is your moderator).

[@yasseryahya4094]

Dear Dr. Gaskin, if I have a third order mediating consturct (reflective-formative-formative), should I follow the same steps as shown in the video ? Thanks in advance.

[@Gaskination]

If it is reflective at the highest level, then yes. But if it is formative at the highest level, you should follow this one instead: kzbin.info/www/bejne/goOxdWB-Ys2Eh9k

[@yasseryahya4094]

Many thanks Dr. Gaskin

[@baokhanh2523]

Prof. Gaskin, how can I interpret the value original sample in mediation test? Thank you!

[@Gaskination]

Original sample is the regression estimate without bootstrapping. Sample mean is the regression estimate average across all bootstrapped samples.

[@baokhanh2523]

@@Gaskination Prof, I have one more very confusing problem. I do mediation analysis based on Baron and Kenny which consists of 4 steps: IV to DV, IV to MV, MV to DV and IV to DV with MV controlling. As the from the first step if IV to DV is insignificant then the mediation role of MV is insignificant. However, as in your video, even the first step is insignificant, you can conclude that it is full mediation? And in the case that IV to DV is insignificant, the rest is significant. can we conclude that MV mediates IV to DV?

[@Gaskination]

@@baokhanh2523 There has been a lot published on mediation (see here for a few: statwiki.gaskination.com/index.php?title=References#Mediation). More recent literature than Barron and Kenny suggests that if the indirect effect is significant, then there is mediation. A non-significant direct effect (when no mediator is present) may indicate that there are multiple competitive effects (see 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.).

[@debanitadasgupta790]

Hi James, Can we have more than one mediation variables from IV to DV ? Eg : IV --> A --> B --> C --> DV How do we see the effect of individual mediators in that case? Thank you...

[@Gaskination]

Yes. This is called serial mediation. You can see the indirect effects of all these in the indirect effects table in smartpls. These are all generated automatically.

[@debanitadasgupta790]

Thank you so so much James... You are a savior ... God bless you ...

[@KamalAsasi]

Dear Prof. James Gaskin. Thanks for the excellent video. I am using SmartPLS (v.2.3.7). There is an option for specific indirect effects. But it is not covering all indirect effects in the model. Any advice? Thanks.

[@Gaskination]

Make sure to run the PLS algorithm, rather than the consistent algorithm. Sometimes the consistent one doesn't produce certain estimates.

[@KamalAsasi]

Thank you dear Prof, James Gaskin. I did it but I got the same results.

[@Gaskination]

Dang. I'm not sure then. Must be a bug in the software. Hopefully they'll fix it soon.

[@KamalAsasi]

I think so, dear Prof. Thanks.

[@debanitadasgupta790]

Hi James, in order to test mediation, should we not test first the direct effect(IV to DV) without the mediator variable included in the model and then the indirect effect with the mediator variable included? Because as soon as we add the mediator variable, the effects will change... Pls clarify... Thank you

[@Gaskination]

Debanita Dasgupta That is the old way to do it. Barron and Kenny advocated that approach. The new approaches recommend just testing indirect affect. The direct effect can be present if necessary to obtain sufficient model fit.

[@debanitadasgupta790]

James Gaskin .. thanks a lot sir

[@dr.bashirahmad6538]

Is it essential to link IV with DV, if we have serial mediating model (e.g., X ->M1->M2->-Y)? Or we can run the model without linking the X directly with M2 and Y and can draw conclusion about mediation on the basis of indirect and total effects? Thanks in advance.

[@Gaskination]

It is not necessary to link it, particularly if you have good model fit. However, if your model fit is poor, then consider linking it.

[@dr.bashirahmad6538]

Thank you so much Sir.

[@anafauziya3058]

Dear Dr Gaskin, thank for you tutorial video. i have a quetions, can i include control variable in mediation? and if yes, how? Thank you

[@Gaskination]

Yes. Include it just like you would include any other exogenous variable.

[@anafauziya3058]

James Gaskin thank you so much...

[@ansariayesha8161]

Dear DrJames, I ran a path analysis through pls algorithm and got my beta values. Later i ran mediation analysis and the values of mt betas changed. Is this normal? which betas should I ideally quote in my path analysis. The ones before mediation or after?

[@Gaskination]

This is normal and expected. Report both, but rely on the mediation ones for testing the hypotheses. Reporting how values change after adding a mediator is informative.

[@ademkara3800]

Mr. Gaskin, while doing mediation analysis in smartpls, can we get the right result if we do not take into account the dimensions of the independet variable? In other words, assuming it consists 4 dimensions, but if we accept as one dimension and analyze, is the result reliable? Thanks.

[@Gaskination]

If you mean you are using a 2nd order factor with four dimensions, then yes, it will work the same either way. The results will differ slightly due to measurement differences in the two models, but the results should be mostly consistent. If instead your mediator has multiple dimensions (and is formative), then you'll need to follow a two step approach (producing latent variable scores and then testing mediation).

[@ademkara3800]

@@Gaskination Thank you very much Mr. Gaskin, I am very grateful to you. You are a real savior.

[@Gaskination]

@@ademkara3800 Lots of good references here: www.smartpls.com/documentation

[@ademkara3800]

@@Gaskination Mr Gaskin, in smartPLS, the relationship between IV and DV was negative, but the effect turned into positive in the triple structural model i established with the mediating variable. How do we interpret this? Is it a problem?

[@Gaskination]

@@ademkara3800 I don't know what you mean by "triple structural model". As for effects switching signs in mediation, this is usually called competitive mediation. You can refer to: 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.

[@hamidreza4501]

Since PLS calculates the net indirect path value for multiple mediators, what do you think if we run PLS with the path to only one mediator each time and run PLS several times to calculate all indirect path values for all mediators?

[@Gaskination]

SmartPLS 3 added a new feature recently to address this to assess individual (specific) indirect effects: kzbin.info/www/bejne/sGLLZ6SDmLKbjc0

[@hamidreza4501]

Thanks for sharing the link.

[@fazfaz2605]

In my model, paths a and b are significant. The direct effect is also significant. But indirect effect is not significant. What does it mean?

[@Gaskination]

It means there is no mediation.

[@ammaryasir4103]

Respected Mr. JAMES GASKIN, sir, if my direct effects result is negative for one relationship. Does the mediation exist? i am calculation mediation through VIF=a*b/total effects. Please reply me. Thanks in advance.

[@Gaskination]

Mediation can be positive or negative.

[@ammaryasir4103]

@@Gaskination ok sir , thanks, it means if direct effect is negeative, still we can report mediation percentage. Thanks sir.

[@Gaskination]

@@ammaryasir4103 I'm not sure about a percentage. There are a few different kinds of mediation. If your direct effect is negative, but the indirect effect is positive, then this is called competitive mediation. Here is a good article on this: 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.

[@ammaryasir4103]

@@Gaskination ok sir, Thanks , its my honor that you replied me. Thanks again.

[@madrevermeulen7925]

If a direct path/relationship is not significant, do you REMOVE this path then before continuing to, for example, MGA?

[@Gaskination]

It is optional to remove. Removing it creates a degree of freedom, but keeping it allows you to "control for" the effects of that variable. Most would argue to keep it.

[@TheKoaydarren]

Agree with Dr Gaskin.. Also, you have to connect the direct path in order to calculate the VAF (partial or full mediation)

[@wendyng8698]

Hi Dr. Gaskin, I have a question about PLS that I couldn't find the answer anywhere or any paper that has similar problem like mine. Is it ok to email you?

[@Gaskination]

sure. google me to find my email

[@buhlenincube2984]

Hi James Thanks for your easy to follow videos. In my study, I want to add a moderating effect on two constructs (I want to measure the moderating effect of gender on hedonic motivation and effort expectancy) and these two constructs have a direct relationship with intention to use. Now my question is, can I do this with Smart Pls 3 ? Can smartPls be used to measure a UTAUT model. Thanks in advance

[@Gaskination]

Buhleni Ncube Smart PLS 3 can be used to test that model. Use the MGA feature that I show in one of my other videos. Gender would be the grouping variable in this case. kzbin.info/www/bejne/mGSQlaycnax4atU

[@dianajacinto6827]

Hello. Thank you for the videos. I have a problem. My model has no relation to only three variables. How can I do a mediating analysis? thanks

[@Gaskination]

I don't think I understand. Perhaps email me a picture of your model. You can find my email by googling me.

[@ammaryasir4103]

@@Gaskination sir you are so kind to us.

[@debanitadasgupta790]

Dear James. I need ur help. In my model, the indirect effect is significant when i DONOT have a direct path between main IV and DV. But as soon as i add a direct path, the indirect effect becomes non significant. How do i interpret thus behavior?

[@Gaskination]

This means that the mediator is not the reason for the effect of IV on DV (i.e., no mediation).

[@debanitadasgupta790]

@@Gaskination Thank you so much James. I have written u an email, pls can u check. Subject is Help - PLS SEM. Thank you

[@chakitanadiah8206]

Dear Mr.Gaskin, what if my mediation is directonly-nomediation? How do i interpret? Thankyou

[@Gaskination]

Then there is no mediation.

[@chakitanadiah8206]

@@Gaskination can u please explain it to me how could that be no mediation there? and why is the results different from the hypothesis before? Thankyou sir

[@Gaskination]

@@chakitanadiah8206 If the indirect effect is not significant, then there is no mediation. Mediation is the indirect effect. This is like if I tell you I have no dog, and then you ask me what kind of dog it is. There is no dog.

[@chakitanadiah8206]

@@Gaskination ok sir, now i understand, thankyou for your explanation

[@digitaleducationist]

Dear Dr Gaskin, can you please tell me ..how we can interpret the results if the IV has 3 formative dimensions!!

[@Gaskination]

Adnan Sheikh Same interpretation as with any other independent variable. Just look at the regression weight between independent variable and the dependent variable. Or if it is mediation, look at the indirect effect. The p-value threshold is the same as with any other regression coefficient.

[@digitaleducationist]

James Gaskin Sir, actually when i see the indirect effect it is showing the formative dimensions as IV and making the high order as mediation and actual mediation after the high order... i just want the result of high order to mediation and mediation to DV (indirect) because high order IV already representing the dimensions...kindly guide in this case

[@Gaskination]

In this case, you can look at the specific indirect effects: kzbin.info/www/bejne/sGLLZ6SDmLKbjc0

[@snk2818]

Hello James, Thanks for your videos. I was looking at dome papers that test endogeniety before SEM.. can you guide me how can we do that?

[@Gaskination]

Endogeneity is when another variable causes multiple other variables to be apparently correlated, even if their correlation is actually due to this other third variable. So, statistically, this could be tested with a simple correlation analysis, and could be prevented by ensuring model fit (in the structural model). Theoretically, it can also be prevented (sort of) by including a nomologically complete model (i.e., not neglecting important covariates and control variables). This is why a literature review prior to study design is so important - so that no critical variables are neglected.

[@snk2818]

When you said ‘could be prevented by ensuring model fit’ does that mean we dont really need to test for endogeniety before SEM if we have a complete model? Just to confirm

[@snk2818]

I found a paper where the authors tested for endogeniety before SEM in smartpls.. the paper is by Upadhaya, Munir, Blount & Su (2018)- Journal of Business Research

[@Gaskination]

@@snk2818 thanks!

[@Gaskination]

@@snk2818 MOdel fit implies that all major relationships in the model are accounted for. So, if there was some present variable that accounted for a relationship, but wasn't being estimated, then model fit would be poor.

[@ss-ci8tc]

i have two groups with different sample sizes. i. e supervisors 100 and employees 300.... both are evaluating each other with one mediating variable. and there is two dependent variables...... how i analyze in SmartPLS?

[@Gaskination]

If the theory is about moderated mediation, then you'll want to compare the indirect effects in an MGA. If the theory is about moderation and mediation (separately), then check mediation with the whole sample, and moderation by doing MGA.

[@isaackosi2889]

Hello DR. Gastin, How do you handle singular matrix error in SmartPLS?

[@Gaskination]

I'm not sure. I've never run into that issue. My first approach would be do to some data screening prior to bringing it into smartpls.

[@abideenadewale7509]

The problem may be due to a very high correlation among some variables. Check the VIF for any multicollinearity issue.

[@dr.mohammadmasudurrahman7401]

Does it happen for the multi group analysis for moderating variable? If it is then I can share my experience.

[@yogapower6988]

Thank you so much for doing the SmartPLS series! Chapter 7 of Hair et al. (2017) "P Primer on Partial Least Squares Structural Equation Modeling" talks about how to split a pooled indirect effect for multiple mediations with bootstrapping. Perhaps you can make a video on that?

[@Gaskination]

thanks! I'll take a look and see if I can make a video for it.

[@RoamingEveryday]

Can you please send me this book if you have it in soft form? My email id is [e-mail]

[@Gaskination]

I don't have a copy.

[@RoamingEveryday]

Oh, I'm sorry I was asking this from "Yoga power" since he/she referred that book. But hey thanks for your concern.

[@mahmad1170]

Dear prof how can i justify negative mediation? Can you please share any article.

[@Gaskination]

Mediation is just the multiplication of a series of direct effects. If there are an odd number of negative effects in that series, then the resulting indirect effect will be negative. So, it's just math. No special references needed.

[@jocaaan]

thanks a lot !

[@thikaartika6082]

thank you for the video. I have a problem. bdw.. I don't understand analyze for sem. 😔 what is this mediation. I don't know.

[@Gaskination]

The mediator is the reason (or part of the reason) for the effect the IV has on the DV. Here are some resources: statwiki.kolobkreations.com/index.php?title=Structural_Equation_Modeling#Mediation Or, a simple video here: kzbin.info/www/bejne/jYvVYp2igM6SqZI

[@dynomowong]

How to use PLS-SEM to run sequential mediation or serial mediation ?

[@Gaskination]

If you have multiple paths and serial paths in the same model, and you wanted to theorize about them separately, then you would probably have to reduce the model to the serial path only. It would be cool to be able to highlight a specific set of paths and run a mediation on that. But that's not an option.

[@prashantchopdar]

Can i perform serial mediation with SmartPLS?Any papers on how to do it?

[@Gaskination]

Yes you can. Just check the indirect effect as you would with regular mediation. The only trick is if you have multiple indirect paths between a single IV and DV (usually with parallel mediators, rather than serial). In this case, there is no good way to do it yet, although I talked with Dr. Ringle and he is planning on implementing this functionality in a subsequent version.

[@prashantchopdar]

Thanks Prof.Gaskin

[@dr.mohammadmasudurrahman7401]

How to calculate Indirect effect for simple mediation?

[@Gaskination]

This video shows how to calculate all indirect effects (regardless of model complexity): kzbin.info/www/bejne/sGLLZ6SDmLKbjc0

[@shaziahasan1722]

Dear Dr. Gaskin, I have some problems with Smart Pls while analysing my thesis data and coludn't get any help from anywhere. So. can I please ask for your mail ID. So that I can send some attachments of models I have run.

[@Gaskination]

shazia hasan if you Google me, you will find my email.

[@shaziahasan1722]

Thank you Dr Gaskin

[@nohox9444]

I think you forgot to calculate the VAF

[@Gaskination]

What is VAF?

[@nohox9444]

Oh hi, cool that you answered. I wanna use that chance and tell you how helpful and great your videos are! Good job! Yesterday I actually read an article about how to calculate partial and full mediation based on Hair et al.'s book (Hair Jr, Joseph F., et al. A primer on partial least squares structural equation modeling (PLS-SEM). Sage Publications, 2016.). They calculate VAF scores to determine the degree of mediation. However, this seems to be outdated and is therefore not supported by smartpls. Instead, smartpls follow Zhao et al.'s article on mediation (Zhao, Xinshu, John G. Lynch Jr, and Qimei Chen. "Reconsidering Baron and Kenny: Myths and truths about mediation analysis." Journal of consumer research 37.2 (2010): 197-206.). They created a new typology of mediations and nonmediations. Your video is a great entry point for everyone who is interested in how to calculate mediation, however, to report mediation correctly in a scientific paper I strongly recommend to read some literature about it. Because what you actually did in the video was to show an "Indirect-only mediation", which is great however you called it a full mediation. But it would just be a full mediation if the VAF score is greater than 80%. Please correct me if I'm wrong, I'm a newbie in PLS-SEM

[@Gaskination]

Got it. yes. You are correct. According to Zhou, it is indirect-only. Full mediation is the old language from Barron and Kenny.