SmartPLS 3 Running and Interpreting a causal model

[@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!

[@goranpavlovic4289]

Your tutorials for SmartPls are great

[@Prageethroshan]

Great video. Very helpful.

[@barrieahmed3233]

Mr. James Gaskin, thank you very much for your good videos on PLS analysis... I've a question: in a thesis report if more than half of the hypotheses are insignificant how can I make them significant before finishing the report?

[@Gaskination]

Here's a post I've written about this issue: gaskination.com/forum/discussion/6/why-i-found-no-significant-relationships-and-all-my-hypotheses-failed#latest

[@barrieahmed3233]

@@Gaskination Thank you.

[@Helwooo]

Great video

[@kidovatov]

Hi Dearest Mr. James Gaskin. I have a lot of problems with my SEM SMART PLS3. The numbers I use are very bad. I have to drop and repeat the calculation for the PLS Algorithms. The other problem is about Singular Matrix Problem. I have to drop so many indicators. Please give me some advice to solve this problem. How can I manipulate the numbers in order to get good result without dropping or even avoid Singular Matrix Problem. Thanks

[@Gaskination]

If the data is bad, it cannot be fixed afterward. Garbage in, garbage out. To avoid a singularity matrix, make sure all of your variables have variance (i.e., aren't constant) and make sure any sets of dummy variables omit a reference category.

[@fatimaikram151]

Thanks James for an amazing video. I have negative correlations between variables. Therefore, negative values inthe Fornel and Larcker table as well. Is it acceptable? Overall , square root values inthe diagonal are higher than it correlation coefficients (off-diagonal). Could you please guide. Thanks

[@aichingyap4824]

Hi, James. The model in this video is mediation, should we perform causal analysis for moderation as well?

[@Gaskination]

If your theory involves moderation, then you should test moderation. Here is a video on MGA: kzbin.info/www/bejne/mGSQlaycnax4atU and here's one on interaction: kzbin.info/www/bejne/hp-zl4J9qdFgosU

[@marleyalvory3044]

Wonderful video Dr. Gaskin. I am wondering what is the difference between Warppls and Smartpls? Why some author would prefer to use Warppls instead of Smartpls?

[@Gaskination]

I have not used WarpPLS very much, but my understanding is that WarpPLS allows the estimation of non-linear relationships.

[@marleyalvory3044]

@@Gaskination Oic Thank you Dr. Gaskin :)

[@gingerginger733]

Thanks again, dear Prof. Gaskin. If I use PLS to analyze the causal relationship, is it better to do EFA using SPSS before Consistent PLS - path? And should I use different data set to do two different work?

[@Gaskination]

If all factors are reflective, then an EFA is a good idea, although not required if all sets of measures have already been validated in prior research. If you have access to SPSS and you are familiar with EFA, then it is a quick process to explore the factors anyway. If you opt to skip the EFA, then just make sure to do a good discriminant validity analysis in SmartPLS (using the HTMT method).

[@ahmedgamal2478]

Hey James! What is the right report for hypothesis testing is it reporting the Path coefficients T statistics and P values or the Latent Variable correlations P-Value and T statistics? Its a casual reflective model 1 dependant variable and 2 independent variables

[@Gaskination]

For causal hypotheses, we report the regression weight and t-stat (or p-value). It is also good to report the full correlation matrix if it is available.

[@nullnorth]

Thanks for this video James! I have a question, why are the numbers on some arrows 0.00? what does it mean? thank you :)

[@Gaskination]

I don't see that occurring in the video above. If that is happening in your own model, then it is because there is no relationship between those variables connected by that arrow. This also happens when using the repeated indicator approach.

[@nullnorth]

@@Gaskination yes its happening in only one variable in my model! i have a problem with my questionnaire thats why. should i just report it as it is and mention the error in the limitation? thank you so much

[@Gaskination]

@@nullnorth If the variable is invalid due to the questionnaire, then you should drop that variable and report it (if it is an important part of your model). If it was just a peripheral variable (not very important), then you can drop it and not discuss it at all in your report.

[@suuuuuuuuuuus]

Hi James, What is the Original sample O number stand for? where can I find the coefficient b, where can I find it in bootstrapping results? Can you do more complex mediation models on SMART PLS and how to report them and interrupt them APA style, please. Thank you. you have been really helpful.

[@Gaskination]

I don't know about APA style. As for the O, it stands for Original regression weight (coefficient B).

[@dipeshkh]

@saraR were you able to find answer for your question about APA style? I am searching for the same now and any help would mean a lot.thanks :)

[@suuuuuuuuuuus]

dipesh khanal hey, not really but I found SEM papers in my field that I’m sure uses apa style and I copied there way

[@dipeshkh]

@saraR can you share the paper with me? [e-mail]

[@suuuuuuuuuuus]

dipesh khanal check vacation posts on Facebook: a model for incidental vicarious travel consumption on journal of travel research I’m sure the authors are using apa style Good luck

[@bubbleblower9993]

Hello professor, I've recently started using smartpls for my undergraduate thesis and found that your videos are very helpful. I want to ask though. my endogenous variable is using a single-item construct derived from secondary data (the dependent variable is measured using Tobin's q) and i read at Hair, 2016 that blindfolding is also used for endogenous variables of single-item construct, why is that? and what determines the value of the omission distance D when the exogenous variables in my research also use secondary data?

[@Gaskination]

Sorry for the delayed reply. I've been busy trying to fix the StatWiki. It is fixed now though. As for your question, I'm not sure. I have never used blindfolding before, so I'm not familiar with its nuances and implications. Sorry about that. Best of luck to you!

[@wisehumansaid]

hi Mr Gaskin, can i ask you please if the original sample is the same as standard beta? Is it the original sample which tel us the nature of the effect (positive Vs negative) of one variable towards a dependent other? thank you Sir ^^

[@Gaskination]

Correct. Original sample is the standardized beta without bootstrapping. The Sample Mean is the average standardized beta of all bootstrap estimates.

[@rotekoppryddeglad]

Hi.Thank you for your videos, they are helping so much. I wanted to ask you if I need to perform psl algorithm before Bootstrapping or if I can go straight to Bootstrapping?

[@Gaskination]

You can go straight to bootstrapping.

[@tilagavatisubramaniam2764]

Hi Mr. James for reflective constructs VIF must take from inner value rite, not outer values formative construct to assess the outer value of VIF?

[@Gaskination]

correct.

[@tilagavatisubramaniam2764]

@@Gaskination thanks alot

[@squidink58]

Hi James! Great video. I'm getting t-values of 0.000 for everything. Have I missed something here?

[@Gaskination]

That's definitely not right. That means that nothing has an effect on anything. I would recommend doing an EFA in SPSS first (if you have reflective factors). This may reveal the issues.

[@squidink58]

I turns out that it was a only an issue with the consistent PLS algorithm. The results are very normal looking when using the regular bootstrapping option. I might perform an EFA anyway to see what's happening. However, my Cronbach's were all good

[@Gaskination]

oh good. glad you figured it out.

[@sumedhachauhan]

Measurement and path models were tested on the same sample. Can you please let me know the potential risks of having this action?

[@Gaskination]

I have no problem with using the same dataset for both. I see no risk. However, it is considered more robust and valid to test on a separate sample. But, usually we don't have enough sample size to do that.

[@bizanjo]

Hello Dr. , What is difference between running PLS algorithm using path analysis or factor analysis? When do we need to run factor analysis ( I mean for which calculations) and when do we need path analysis? Does this has any relationship in analyzing measurement model or structural model?

[@Gaskination]

Use Path for structural model assessment. Use Factor for measurement model assessment.

[@bizanjo]

Thank you very much Dr. Really appreciate your response

[@bizanjo]

I just have one more query. Is it necessary to perform EFA (Exploratory factor analysis) before using SmartPLS? I Have read that in assessment of measurement model you are already performing factor analysis( CFA). So do i need to run EFA or can start with analysis of my measurement model?

[@Gaskination]

It is not required, although I still find it helpful to determine discriminant validity issues. If you are not able to resolve the factor analysis in SmartPLS, you can go back to EFA in SPSS.

[@bizanjo]

Thank you very much Dr.

[@wlidster]

Mr. Gaskin. I am trying to use SmartPLS3 for UTAUT. It looks like in your videos that you are using something similiar. But, the ultimate DV -- "Use" --- doesn't have indicators in the instrument....unless you take the question that asks whether a person intends or actually has adopted. What do you think?

[@Gaskination]

I have similar data, but on many different variables, not all from UTAUT.

[@egonseldeslachts5454]

Hello William and James, I know your question dates from over 2 years ago, but I am in the same situation, assessing the UTAUT with SmartPLS 3 and wondering what to do with the variable Use, since it doesn't really have any indicators. I did measure Use using a question which asked how much the person used the technology. Can I just treat use as a latent construct and connect it to that one indicator? I actually have the same question about moderating variables like age and gender in the UTAUT model, do I just connect them to one indicator? If you happen to remember how you solved it, I would love to know! PS James, thank you so much for your video's, they are excellent and very helpful for a novice like me. Keep doing what you're doing!

[@lorenzoduchi7871]

Hi, I wanted to ask you why I see different path coefficient after having used PLS algorithm vs Bootstrapping. I thought they should point to the same as the Bootstrapping is checking the significance of the path coefficients found with the PLS algorithm. But now the results are quite different (e.g. correlation from -.285 to -.15): WHY?

[@Gaskination]

it is natural for them to be different. The bootstrap runs the PLS algorithm 2000 times (or however many you specify), but each time it runs, it replaces some values with average values. Naturally this changes the estimates. The more reliable estimate is the bootstrap estimate.

[@ghanimsaqib6101]

Can I use regular bootstrapping instead of consistent bootstrapping as all of my constructs fails in consistent bootstrapping but passes in regular bootstrapping. What should I do? My model has all the reflective measurements.

[@Gaskination]

Consistent bootstrapping has some known issues, so you would be justified in using regular bootstrapping. The problem with regular bootstrapping is that it sometimes inflates estimates when all factors are reflective.

[@gunawanprabowo8183]

what is number on indicator in model, when we run bootstraping? in this conteks on the pict model "skillacq_1= 3.948" "skillacq2=7.282". thankyou @james

[@Gaskination]

These are t-statistics (critical ratio). They indicate the extent to which you were able to reduce error and therefore, the significance of the path. Anything over 1.96 is considered "significant" at the 95% confidence level.

[@gunawanprabowo8183]

thanks for respons james...but thats i thinks no t statistisc number... the number each indicator not in path beetween variable. thankyou james. what is the interpretation for that number? minute 6:54

[@Gaskination]

@@gunawanprabowo8183 yes, those are also t-statistics indicating the significance of the coefficient for that indicator with that factor.

[@drkeyurnayak7812]

Very Nice Examples, how to get a datafile for this tutorial.

[@Gaskination]

Most of my video datasets are available on the homepage of statwiki.

[@maryyy5551]

Hey Mr. Gaskin, It's me again ;) I am running a bootstrapping for my rather complex model that contains various interaction effects. The bootstrapping takes really long (it said 30,000 seconds). I already changed the calculation method for the interaction effects from "orthogonalizing approach" to a different approach, but the time taken for the bootstrapping does not decrease. What can I do? Thanks in advance for an answer!

[@Gaskination]

You can reduce the number of bootstrap samples, and you can use the two-step approach (fewer parameters).

[@drkeyurnayak7812]

Thnaks Sir

[@NZeDt]

Hey dear James, PLZ what is the difference between Pls Algorithm and Consistent Pls Algorithm; Bootstrapping and Consistent Pls Bootstrapping ? under which conditions we should use one vis a vis the other?

[@Gaskination]

Use PLSc if you have a model with all reflective factors. Use regular PLS algorithm for any model that has even one formative factor.

[@dr.villyabraham]

Hi James, I'm writing a paper based on a model I;ve created using smart pls 3.0. I'd like to report model fit indices such as GFI, RMSEA, etc. but I don't see them in the report output. Is there an option for me to set the type of indices I'd like to be included in the report? Thanks!

[@Gaskination]

GFI etc are not relevant to PLS models. Here is a video on model fit in PLS: kzbin.info/www/bejne/gqGWo2h8YsqJedU

[@wafajaouadi6566]

Hi, is it important to put all indicators ( f2 Q2...) in thesis to estimate structural model

[@Gaskination]

There are a lot of validity indicators that could be included. It is probably not necessary to include them all, but I would recommend including the two you mention.

[@wafajaouadi6566]

@@Gaskination thank you Mr james you help me a lot no words to thank you

[@suuuuuuuuuuus]

Hi do we judge the model fit with the saturated SRMR saturated model or estimated model? because for me one of them is above .8 and one is less than .8? Thanks,

[@Gaskination]

The estimated model is your model. The saturated model removes all degrees of freedom (i.e., implies all possible parameters are being estimated). Therefore, it is more accurate to use the estimated model.

[@tasneemsaifuddin5862]

I am a researcher and I want to know that if I have to compare two models like model A and model B and to see which is better how can I check it in smart pls? Can I simply compare on the basics of T statistics of both models Waiting for your reply

[@Gaskination]

Usually model fit is used. However, SmartPLS does not rely on fit statistics. So, t-stats, effect sizes (f2), predictive values (Q2, R2) and regression weights are the best approach.

[@tasneemsaifuddin5862]

@@Gaskination thank you so much for your reply

[@jalalahamed417]

Dear Sir: I was running a causal model in SmartPLS 3 (student version). All the latent variables are reflective. Hence, I was trying to run the consistent PLS algorithm. It produces some strange result. Some factor loadings are more than 1, while some of the paths result are with n/a. Instead, when I run the PLS algorithm (without consistent), the results look as expected. Same for the bootstrapping. Now, my question is, in this case, if I run the PLS algorithm (without consistent); will it be methodologically correct?

[@Gaskination]

I have experienced the same issues. I'm not sure what causes these errors. If you can't obtain AMOS, then just use SmartPLS for it. If you end up with a purist reviewer, you may end up needing to switch to a covariance-based software.

[@jalalahamed417]

@@Gaskination Thank you very much

[@najamulhassan4852]

Hi James thanks for the awesome tutorial, could you please tell me? how can I incorporate 4 IV and 1 DV with time series data of 18 years only in smart pls 3.0, I'm confused, if it can be !!

[@Gaskination]

SmartPLS is not good with time series data. You would need to use another software for that.

[@ahmed3820]

Doctor, i have around 40 responses to a face to face interview survey. I used pls for my model but some constructs failed the convergent validity test ave, cr and alpha. Is there any way to avoid this , maybe bcz it is an interview?!!

[@Gaskination]

interviews rarely produce reflective latent factors. So, make sure you are analyzing your data with the proper method.

[@ahmed3820]

@@Gaskination do you advice any method doctor, I feel stuck 😔 Just consider a 45 response , forget about the interviews is there any technique that can be used other than sem

[@Gaskination]

@@ahmed3820 If those interviews produced meaningful numbers, then you can always do first generation statistics like t-tests, anova, regression, correlations, etc.

[@alialshebami8408]

Is there a correlation in Smartpls? how to calculate it?

[@Gaskination]

You can choose to connect all latent variables (see video above at 1:32) as an option when running the analysis. It is a checkbox on the setup tab. This should then produce a correlation matrix in the output under the label "discriminant validity".

[@alialshebami8408]

@@Gaskination thanks

[@SA-fq2qw]

If my instrument scale have two factors and i have to apply cfa then jow can i apply??

[@Gaskination]

Here is a video for CFA in SmartPLS: kzbin.info/www/bejne/gJDIpXqfrMeCpa8

[@saharsarhan]

? Hi, Can I use the total degree of the inner model instead of using single items of the outer model

[@Gaskination]

If you mean you want to use latent variable scores (composite variables) then yes, you can do that. It is just a two-step approach.

[@osohaib111]

What is the difference between causal model, common factors model and composite model?

[@Gaskination]

osohaib111 A causal model is one that has causal paths. This could be a path model or a structural equation model. As long as there are regression lines between constructs. A common factors model is one that includes a common factor connected to all observed variables. A composite model is one that uses latent factor scores instead of latent variables.

[@paulmburu132]

Do you need help with your dissertation? Do you require to implement a complex SEM model? I am ready to help.

[@EPCEXPERT]

Yes, please

[@paulmburu132]

EPCEXPERT I am ready to help you. How can we connect?

[@EPCEXPERT]

Dear Paul Please contact me on [e-mail] Speak soon

[@Aidino123]

@@paulmburu132 Hi Paul, can you possibly send me your email? I need some help with my dissertation.

[@corneliareza6537]

How to know strandard error (STERR) in pls?

[@Gaskination]

Unfortunately, SmartPLS doesn't report Standard Error. However, it does report Standard Deviation. Standard error = Standard deviation divided by the square root of the sample size.

[@corneliareza6537]

James Gaskin standard deviation in path coefficients or total effect?

[@Gaskination]

@@corneliareza6537 It estimates them for both when you bootstrap.

[@RicardoGouveiaRodrigues]

Is your dataset avalialable?

[@Gaskination]

The SPSS version of the dataset is available on the homepage of statwiki. It is the SEM Series dataset.

[@Videos-rw8ys]

KZbin SEM series.sav doesn't match with this video :( The CSV file converted from the SAV won't let the smartPLS to recognize the indicators. Would you possibly upload a csv version of the correct data file here?

[@Gaskination]

drive.google.com/open?id=1PJa_XJWIwK52cw-nd9Uy_bfYEm7NaRwp

[@RicardoGouveiaRodrigues]

Thanks.

[@RicardoGouveiaRodrigues]

I'm afraid this is the same dataset avaliable on statwiki. Would you please provide a new link? Thanks.

[@dr.shabnamnarula4097]

Sir from option of SmartPLS 3 I can identify the redundancy

[@unconsciouscognition7449]

Yes. Here is a book reference about it: books.google.de/books?id=Xn-LCwAAQBAJ&pg=SA5-PA23&lpg=SA5-PA23&dq=smartpls+3+redundancy+analysis&source=bl&ots=sm77vQjOVJ&sig=ACfU3U378AZTm24V5NgeFOQ17yY3RnvX_w&hl=en&sa=X&ved=2ahUKEwjq2p3HubfmAhVssYsKHSP-CgkQ6AEwDHoECAsQAQ#v=onepage&q=smartpls%203%20redundancy%20analysis&f=false

[@azakakmohamed432]

Hello dear teacher, Please I have a problem with the analysis with smartpls 3, it does not work when computing by pls algorithm.

[@Gaskination]

What do you mean by "it does not work"? Do you mean that some values are missing on the model, or the report is incomplete, or an error occurs, or something else? I have never run into any errors in SmartPLS, so I don't know that I would be very helpful in troubleshooting them anyway.

[@azakakmohamed432]

Good afternoon professor, It's settled, just i have changed the language in english and the program has been running. thanks for your help

[@arunpatil5653]

It is recommended not to use SMART PLS if the sample size is greater than 200. Is there any particular reason to use SMART PLS when sample size is greater than 200??.. Sir please share your comments

[@Gaskination]

I have never heard that recommendation.

[@arunpatil5653]

@@Gaskination So any particular reason to use Smart PLS if sample size is greater then 200?

[@Gaskination]

​@@arunpatil5653 No. The only sample-size criterion that influences your software decision should be if the sample size is low. SmartPLS converges more easily than covariance-based methods. This means it can handle smaller sample sizes. However, the error associated with small sample sizes is still a concern.

[@zackarnie]

Hi,I wanted to ask that where I can get this software Smart PLS from?

[@Gaskination]

if you google it, here is the first result you'll see: www.smartpls.com/

[@NZeDt]

Hey dear James, PLZ what is the difference between Pls Algorithm and Consistent Pls Algorithm; Bootstrapping and Consistent Pls Bootstrapping ? under which conditions we should use one vis a vis the other?

[@Gaskination]

Use PLSc if you have a model with all reflective factors. Use regular PLS algorithm for any model that has even one formative factor.