I miss your new video, i hope i will be about some new features of PLS-SEM using smartpls, thank for sharing with us. You are a great and kind professor
Hello, Dr. James Gaskin. 1st I would like to thank you for your valuable inputs in all regards. I am watching all of your videos on SmartPLS 3 nowadays. But, I missed something in your videos that are “referencing.” I think it will be more convenient, helpful, and beneficial for viewers if you added a reference in your videos. As, when I was watching your video on “Model Fit SmartPLS 3” you mentioned that the value of SRMR should be less than “.08.” (but who said that). As in research, we need some references to justify our arguments. So, it's my opinion it may be/may not be right. Hopefully, you will consider it. Thanks Regards
Hello Dr. James, I am glad to hear the first few lines of the video. If you could guide me in one problem in my model. In a reflective -reflective second order model I am getting NFI value as n/a.. what could be possible reason for such result. Also Chi square statistics in the result is being shaown as infinite and SRMR value 0.148...However...all other measurement model and structural model values are being met. Can I go ahead with model? I am not getting any Literature on this kind of error. Shall be grateful for your guidance.
gday Dr James. I've obtained SRMR values 0.081 and 0.088. its pretty worrysome already. but i get more worried when the chi square showed results of 'infinite' and the d_G and NFI showed results of n/a. Im kinda started to thinking im gonna have a big issue here. is there any way this situation can be explained? Thanks again Dr.
I would not trust those results. PLSc is still a little buggy I think... I recommend instead just validating the factors in other ways. Factors that demonstrate convergent and discriminant validity are likely to be part of a good measurement model. In the structural model, you can assess multiple models to show how the R-square changes. If it changes substantially when adding a path that was not originally theorized, then this is a good indication that structural model fit is not ideal for your original model.
Hello Dr.James- thank you so much for these tutorial, extremely helpful and I learned from you a lot. Dear Dr.James I got result after the model estimated based smartpls but some results seem incoherent. The numbers on the arrows, although they express different things (path coefficient versus correlation t-value), e.g., Trust->Adoption has path coefficient equal to 0.429 and a t-value equal to 3.836, while Compatibility->Adoption has a slightly smaller path coefficient (0.441) but the t-value is almost 50% bigger (5.412). pealse, can you provide explanation?
Hi James. Even reflective measurement may include indicators with relatively low loadings. Removing those indicators may thus negatively affect the type of fit you are referring to in this video. WarpPLS implements factor-based algorithms, which tend to improve indicator correlation matrix fit indices. See the videos linked below: ruclips.net/video/PvXuD5COezU/видео.html ruclips.net/video/YutkhEPW-CE/видео.html
Thanks! I've been thinking about learning WarpPLS lately. I was even going to include it in my SEM Boot Camp next week, but then ran out of time to learn it before then... Maybe I'll figure it out over the summer.
Henseler, J., & Sarstedt, M. (2013). Goodness-of-fit indices for partial least squares path modeling. Computational Statistics, 28, 565-580. doi.org/10.1007/S00180-012-0317-1.
Hi Dr. Gaskin, thank you for the great video. I do have a question regarding the model fit for 2nd order factors in SmartPls 3. When should we calculate the SRMR? After plugging Latent Variable Scores into a follow up path model?
If you want to test model fit (which is debatable in PLS models), then it should be tested with models you are using for hypothesis testing. So, if you are testing a hypothesis with the LVS model, then that is the model that should have model fit evaluated.
Hi James- thank you so much for these videos, extremely helpful. I was wondering if it is possible to use these model fit indices to do model comparison? So say I had two different path theories, and I wanted to test which one best fits the data? Is this something that’s often done or do people stick to one model? Many thanks for any advice
Georgie Powell model comparison can be done with model fit. A common approach uses the “delta CFI”. If the CFI is more than 0.01 different across models, then the models are different.
I read that in CB-SEM it is common to make modifications to the proposed model so that it achieves model fit, enhancing the conslusions of the research. Since model fit is not well developed for PLS-SEM (and therefore I decided not to use it in my research), that means that all the conclusions of a study with PLS-SEM derive only from measures like R2 and Q2, right? Is it normal to take the initiative to change the proposed model so that it achieves a higher R2 in PLS-SEM for instance? Thank you for your videos and help!
Hi James, I see you added the relationship from playful to useful for better model fit. For the formative model, if I added relationships to improve R-square, do I need to change the conceptual framework in the dissertation accordingly? Or if it's related conceptually, this is fine and no need to change? Thanks!
Hi Prof Gaskin, I have one major problem with my analysis. I have three IVs (all reflective constructs) pointing to one DV (reflective). Since my model consists of all reflective factors, so I use ConsistentPLS algorithm. However, one of the IVs showed an positive relationship-positive path coefficient (instead of my proposed relationship, which is a negative one). I went through deleting some indicators (since they are less than 0.4 and to meet the requirements of convergent and discriminant validity). However, I can see while deleting some indicators, the path coefficient change from positive to negative gradually. Weirdly speaking, after I have done enough deletion of the indicators of the said IV, it appears to show a negative relationship (my proposed relationship). Could you please advice me on this matter? Interestingly, I tried to conduct my analysis with normal PLS algorithm and the said IV is in line with the proposed relationship with DV(negative). I wonder what could cause this huge difference between using PLS algorithm and a ConsistentPLS algorithm? and would appreciate if you could provide me some solutions on this.Thank you Prof Gaskin.
This can happen if you are using a multidimensional factor or if you have some reverse-coded items in your set of indicators. Typically reverse-coded items, even if re-reversed, have lower loadings and contain much more error than positively-coded items.
@@Gaskination Thank you very much for your input. Yes, they are all unidimensional constructs but do have a number of reversed-coded items. I just came to realise that I have DELETED more than 50% indicators of the overall indicators, original 34 items now left 11 items. Original , Now (AVE, CR, NFI, SRMR improved) 6----------2 8----------3 10-------3 10-------3 but they are (all reflective), is that acceptable? Do I have any literature to support this?My rationale for this action is to improve Cronbach alpha, composite reliability, and AVE. I also did this to improve my overall MODEL FIT. Besides, one of the reviewers did ask me about the issue for dropping so many items especially from well-established scales. My justification for this is that unlike CB-SEM which prefers using more items per construct to improve reliability and measurement model fit, VB-SEM (PLS-SEM) is not restricted with the minimum number of items per construct even can have one indicator for each construct. Do you think this is a good justification?
@@joelyap2799 I would also be concerned if these are established scales. Make sure you have re-reversed them if they are reversed (i.e., subtract values from 1+scale size). You might also try running these through an EFA in SPSS if they are all reflective. This will reveal the true issues.
James: Can you tell me where to find the degree of freedom? And how do you interpret the value of d_ULS and d_G? what are the threshold for these two values? Thanks so much.
Hello Dr. Gaskin. Thank you for your excellent tutorial on the model fit. Please I have a question. It is right to have the saturated and estimated models in the model fit using Smart PLS 3, having the same values , for example SRMR have same value for saturated model and estimated model. Thank you.
I actually prefer covariance based methods, instead of PLS. Most would agree that CB-SEM allows for more rigorous testing of the measurement model when the factors are reflective.
Hello, Dr. James Gaskin. Thank you so much for your demonstration. I have one question. Could you let me know how to lower the SRMR value (0.201) of estimated model in model fit, but 0.028 in saturated model? Of course, I will check the data again.
It could be due to a missing causal path between latent variables. Or it could be due to factor misspecification (as reflective instead of formative, or the other way around). Or it could be due to poor factor convergent validity.
Dear Mr. Gaskin, thank you for your video! I am working with the book Hair et al. (2017) - A Primer on partial least square structural equation modeling. It says that the SRMR should not be used for PLS-SEM and also does not mention the NFI (p.193). Is there a source that says they can be used for PLS-SEM? Thank you in advance.
haha. yes. I've talked with Christian about this. Even though they include model fit measures in SmartPLS 3, they are not happy about it. They included them very reluctantly. Model fit is a covariance-based topic and is not relevant to PLS. So, no, I don't know of any reference for model fit in PLS. Sorry about that. Best of luck to you.
Hello Dr. Gaskin. i remember sometimes ago you had a video on calculating VIF using SmartPLS 3. however, now i cannot find it. Can I ask where it is? Thank you.
Thanks James for the video. My model gives me a pretty decent SRMR of 0.047. However, Chi-square is infinite and NFI is n/a. Am I doing something wrong ???
Model fit is somewhat dubious in PLS still. I've found the estimates are not very consistent. So, if it is not working, I'm not surprised. I would recommend using other ways to validate the model.
@@Gaskination Thanks. Can you please elaborate other ways which can be used for model fit, if I am testing the model in SmartPLS. Thanks in advance and sorry for bothering you.
Hello Professor im actually writting my thesis and i got problems with my models : my sample is Low about 70 questionnaries and i have 7 variabless per model. The model fit i get (within almost 2 indicators per variable) is : SRMR = 0,049 (Estimated 0,101) and NFI = 0,86 (Estimated = 0,79). What you think about these results ? i could not reduce more the indicators.. so it's the maximum that i can obtained (I think).. please need Help :/
I would not trim your model so much to achieve good model fit in PLS. Model fit really is a foreign concept for PLS. You might also consider whether some factors are formative if you are having to trim so much.
How do you interpret the bootstrap confidence interval and p values for SRMR? I have a case where oddly SRMR is 0.08 but the confidence interval is only between 0.01 to 0.04. When I look at the raw bootstrap values, none are even close to 0.08.
That is very odd. Luckily it is odd in the right direction. We want to minimize the SRMR, so those lower values are fine. I'm not sure though why they would be so far from the estimate...
Thank you doctor Gaskin for the valuable videos. I have a question, is it possible to get model fit indexes such as comparative fit index(CFI) and RMSEA in smartpls?
mohammed Al Yami If the p-values are not significant, that can be due to many things. It could be due to poor model fit, poor validities, poor data, poor theory, etc.
Thank you for your great video. I always suggest my students to use your vidoes for their projects and thesis writing. Do you have any academic article written using PLS SEM to use as a reference on how to report results? Thank you.
Thanks! Here is an older article: Lowry, P. B., & Gaskin, J. (2014). Partial least squares (PLS) structural equation modeling (SEM) for building and testing behavioral causal theory: When to choose it and how to use it. IEEE Transactions on Professional Communication, 57(2), 123-146.
hi prof. please i was wandering if my model has 1 IV and 1 DV(with two dimensions) can it be possible to match the IV to the dimensions of the DV and at the same time the DV itself
I'm pretty sure SmartPLS will allow it. Although, it may be hard to interpret. I would recommend testing the model with the DV as 2nd order, and then retesting it with the DV as separate dimensions.
Hi James, thanks for the excellent videos on SmartPLS, can you please let me know how to check the correlation matrix in SmartPLS? and where to find it if it is available ...? if not, what is the answer for somebody ask you about correlation matrix? Appreciate your reply ... Thanks NA
It is best to do everything you can to avoid removing items. So, I would recommend re-reversing the values of those reversed items so that they load positively on the factor. If they are still very low loadings, then I might recommend breaking this into two dimensions and then connect them to a 2nd order factor. If still no good, then yes, it is okay to lose the three.
Goodness of fit is built on the assumption that there is little difference between the observed covariance matrix and the proposed covariance matrix (implied by a model). However, PLS is not a covariance-based method, so a covariance-based measure (like GOF) is fundamentally misaligned with the method.
Only sort of. It's really for all reflective models. But it's really only for covariance-based SEM (not PLS). So, whatever fit measures you observe, take lightly.
If indirect path is significant, but direct paths is not, then full mediation. If indirect path is significant, and direct paths is significant, then partial mediation.
@@juanrizalsaari4635 AMOS is a more appropriate software for models that do not have formative constructs. It uses a covariance-based (maximum likelihood) approach, which is better suited to reflective models. The covariance-based approach is generally more accepted than PLS.
@@juanrizalsaari4635 If you have any formative factors, then PLS is best. If you only have reflective (no formative), then AMOS (or other covariance based approach) is best.
Sir, which model is checked for SRMR (saturated or estimated)? SRMR of saturated model is 0.08 but SRMR of estimated model is o.1001.... is it acceptable? or can you please give me any guideline to understand this software
SRMR should be less than 0.800, but model fit is not totally relevant to a formative model. Joseph Hair and Christian Ringle would probaby say model fit is not the aim of PLS. As for line thickness, I'm not sure. I've never tried to play with the line formatting.
Dear sir, how we have to make factor analysis in case of formative factors and when measured variables or indicators are having cross loading problem (in AMOS or PLS). Kindly suggest How we have to plan removal of indicators in case of formative factors if all are of significance as indicators are non interchangeable. Kindly throw some light in such type of problems. Also if possible kindly suggest the video for messy EFA specially for formative factors.
@@Gaskination Sir in formative factor analysis how we have to extract the factors and associated indicators ? Do we have to use same approach as that for reflective factor but in that indicators are correlated while it is not so in formative case. or will we direct have to build our own factor model then simply test the model ? But this last approach still not give outcome of maximum worth for industrial executives use or interpretation as desired by researcher. So Kindly suggest the method of extracting the factors and associated indicators in formative factors for maximum worth of the model. Also what is the meaning of negative factor weight of indicator on formative factor as we are asking the assertive question in the survey questionnaire. Regards
@@ArtistHridya_2023 I'm not sure I fully understand. Here is a video on how to extract factor scores (for formative and reflective): ruclips.net/video/LRND-H-hQQw/видео.html (I've queued the link to the right time point). As for a negative factor weight, it just means that the indicator is inversely correlated with the other indicators. Make sure it is not a reverse-coded question. If it is, then re-reverse the values (so, if on 5-point Likert scale, this means subtract values from 6).
@@Gaskination Dear sir, May I have any video on messy EFA for FORMATIVE FACTORS for better model planning approach for formative model building? Also what is the meaning of path value coefficient (in measurement model path and structural model path) if we do formative factor analysis in AMOS (as it does not seem factor weights and regression weight respectively, in AMOS as it comes in smartpls)?
@@ArtistHridya_2023 I don't have a video for a messy factor analysis with formative factors... Sorry about that. In AMOS, they are all regression weights. However, I do not recommend formative modeling in AMOS. It was not designed for this.
Hello Mr. James Gaskin. I have been watching your tutorials because I am unable to get reliability of my data in PLS, if you can help me it would be very kind of you. I need help. Please.
I miss your new video, i hope i will be about some new features of PLS-SEM using smartpls, thank for sharing with us.
You are a great and kind professor
True
Hello, Dr. James Gaskin. 1st I would like to thank you for your valuable inputs in all regards. I am watching all of your videos on SmartPLS 3 nowadays. But, I missed something in your videos that are “referencing.” I think it will be more convenient, helpful, and beneficial for viewers if you added a reference in your videos. As, when I was watching your video on “Model Fit SmartPLS 3” you mentioned that the value of SRMR should be less than “.08.” (but who said that). As in research, we need some references to justify our arguments. So, it's my opinion it may be/may not be right. Hopefully, you will consider it. Thanks
Regards
I definitely agree. Here is a list: statwiki.kolobkreations.com/index.php?title=References
Hello Dr. James, I am glad to hear the first few lines of the video. If you could guide me in one problem in my model. In a reflective -reflective second order model I am getting NFI value as n/a.. what could be possible reason for such result. Also Chi square statistics in the result is being shaown as infinite and SRMR value 0.148...However...all other measurement model and structural model values are being met. Can I go ahead with model?
I am not getting any Literature on this kind of error. Shall be grateful for your guidance.
Sometimes PLSc gives inexplicable results. I would recommend to ignore model fit in this case.
Thanks a lot Dr. James for your prompt response.
gday Dr James. I've obtained SRMR values 0.081 and 0.088. its pretty worrysome already. but i get more worried when the chi square showed results of 'infinite' and the d_G and NFI showed results of n/a. Im kinda started to thinking im gonna have a big issue here. is there any way this situation can be explained? Thanks again Dr.
I would not trust those results. PLSc is still a little buggy I think... I recommend instead just validating the factors in other ways. Factors that demonstrate convergent and discriminant validity are likely to be part of a good measurement model. In the structural model, you can assess multiple models to show how the R-square changes. If it changes substantially when adding a path that was not originally theorized, then this is a good indication that structural model fit is not ideal for your original model.
Hello Dr.James- thank you so much for these tutorial, extremely helpful and I learned from you a lot.
Dear Dr.James
I got result after the model estimated based smartpls but some results seem incoherent. The numbers on the arrows, although they express different things (path coefficient versus correlation t-value), e.g., Trust->Adoption has path coefficient equal to 0.429 and a t-value equal to 3.836, while Compatibility->Adoption has a slightly smaller path coefficient (0.441) but the t-value is almost 50% bigger (5.412).
pealse, can you provide explanation?
That is not unusual at all. That is perfectly consistent. The t-value and the path coefficient do not have to move together perfectly.
Hi James. Even reflective measurement may include indicators with relatively low loadings. Removing those indicators may thus negatively affect the type of fit you are referring to in this video. WarpPLS implements factor-based algorithms, which tend to improve indicator correlation matrix fit indices. See the videos linked below:
ruclips.net/video/PvXuD5COezU/видео.html
ruclips.net/video/YutkhEPW-CE/видео.html
The other video on the indicator correlation matrix fit indices generated by WarpPLS: ruclips.net/video/YutkhEPW-CE/видео.html
Thanks! I've been thinking about learning WarpPLS lately. I was even going to include it in my SEM Boot Camp next week, but then ran out of time to learn it before then... Maybe I'll figure it out over the summer.
If I want to write in my paper that it's unnecessary to assess model fit in PLS-SEM, which reference can I cite? Thanks
Henseler, J., & Sarstedt, M. (2013). Goodness-of-fit indices for partial least squares path modeling. Computational Statistics, 28, 565-580. doi.org/10.1007/S00180-012-0317-1.
Hi Dr. Gaskin, thank you for the great video. I do have a question regarding the model fit for 2nd order factors in SmartPls 3. When should we calculate the SRMR? After plugging Latent Variable Scores into a follow up path model?
If you want to test model fit (which is debatable in PLS models), then it should be tested with models you are using for hypothesis testing. So, if you are testing a hypothesis with the LVS model, then that is the model that should have model fit evaluated.
@@Gaskination Thanks very much.
Hi James- thank you so much for these videos, extremely helpful. I was wondering if it is possible to use these model fit indices to do model comparison? So say I had two different path theories, and I wanted to test which one best fits the data? Is this something that’s often done or do people stick to one model? Many thanks for any advice
Georgie Powell model comparison can be done with model fit. A common approach uses the “delta CFI”. If the CFI is more than 0.01 different across models, then the models are different.
I read that in CB-SEM it is common to make modifications to the proposed model so that it achieves model fit, enhancing the conslusions of the research. Since model fit is not well developed for PLS-SEM (and therefore I decided not to use it in my research), that means that all the conclusions of a study with PLS-SEM derive only from measures like R2 and Q2, right? Is it normal to take the initiative to change the proposed model so that it achieves a higher R2 in PLS-SEM for instance?
Thank you for your videos and help!
It is common to test alternative models (where paths are added or removed) to see if predictive quality improves.
Hi James, I see you added the relationship from playful to useful for better model fit. For the formative model, if I added relationships to improve R-square, do I need to change the conceptual framework in the dissertation accordingly? Or if it's related conceptually, this is fine and no need to change? Thanks!
If you add a path, you need to explain why it was added. This is usually done as post-hoc theorizing, rather than revising the original theory.
Hi Prof Gaskin, I have one major problem with my analysis. I have three IVs (all reflective constructs) pointing to one DV (reflective). Since my model consists of all reflective factors, so I use ConsistentPLS algorithm. However, one of the IVs showed an positive relationship-positive path coefficient (instead of my proposed relationship, which is a negative one). I went through deleting some indicators (since they are less than 0.4 and to meet the requirements of convergent and discriminant validity). However, I can see while deleting some indicators, the path coefficient change from positive to negative gradually. Weirdly speaking, after I have done enough deletion of the indicators of the said IV, it appears to show a negative relationship (my proposed relationship). Could you please advice me on this matter? Interestingly, I tried to conduct my analysis with normal PLS algorithm and the said IV is in line with the proposed relationship with DV(negative). I wonder what could cause this huge difference between using PLS algorithm and a ConsistentPLS algorithm? and would appreciate if you could provide me some solutions on this.Thank you Prof Gaskin.
This can happen if you are using a multidimensional factor or if you have some reverse-coded items in your set of indicators. Typically reverse-coded items, even if re-reversed, have lower loadings and contain much more error than positively-coded items.
@@Gaskination Thank you very much for your input. Yes, they are all unidimensional constructs but do have a number of reversed-coded items.
I just came to realise that I have DELETED more than 50% indicators of the overall indicators, original 34 items now left 11 items.
Original , Now (AVE, CR, NFI, SRMR improved)
6----------2
8----------3
10-------3
10-------3
but they are (all reflective), is that acceptable? Do I have any literature to support this?My rationale for this action is to improve Cronbach alpha, composite reliability, and AVE. I also did this to improve my overall MODEL FIT.
Besides, one of the reviewers did ask me about the issue for dropping so many items especially from well-established scales. My justification for this is that unlike CB-SEM which prefers using more items per construct to improve reliability and measurement model fit, VB-SEM (PLS-SEM) is not restricted with the minimum number of items per construct even can have one indicator for each construct.
Do you think this is a good justification?
@@joelyap2799 I would also be concerned if these are established scales. Make sure you have re-reversed them if they are reversed (i.e., subtract values from 1+scale size). You might also try running these through an EFA in SPSS if they are all reflective. This will reveal the true issues.
James: Can you tell me where to find the degree of freedom? And how do you interpret the value of d_ULS and d_G? what are the threshold for these two values? Thanks so much.
No DF because no covariance matrix. forum.smartpls.com/viewtopic.php?t=15945
Hello Dr. Gaskin. Thank you for your excellent tutorial on the model fit. Please I have a question. It is right to have the saturated and estimated models in the model fit using Smart PLS 3, having the same values , for example SRMR have same value for saturated model and estimated model. Thank you.
It is possible, especially if your factors have very few indicators.
Hi, Dr. james this PLS-SEM is more preferably for conformatory and exploratory factor analysis than others, i need your suggestion with great excuse?
I actually prefer covariance based methods, instead of PLS. Most would agree that CB-SEM allows for more rigorous testing of the measurement model when the factors are reflective.
Hello, Dr. James Gaskin. Thank you so much for your demonstration. I have one question. Could you let me know how to lower the SRMR value (0.201) of estimated model in model fit, but 0.028 in saturated model? Of course, I will check the data again.
It could be due to a missing causal path between latent variables. Or it could be due to factor misspecification (as reflective instead of formative, or the other way around). Or it could be due to poor factor convergent validity.
Thank you so much for your comments. The other model fits were very good. So, after I revised the model and got good model fit.
Dear Mr. Gaskin, thank you for your video! I am working with the book Hair et al. (2017) - A Primer on partial least square structural equation modeling. It says that the SRMR should not be used for PLS-SEM and also does not mention the NFI (p.193). Is there a source that says they can be used for PLS-SEM? Thank you in advance.
haha. yes. I've talked with Christian about this. Even though they include model fit measures in SmartPLS 3, they are not happy about it. They included them very reluctantly. Model fit is a covariance-based topic and is not relevant to PLS. So, no, I don't know of any reference for model fit in PLS. Sorry about that. Best of luck to you.
Thank you!
Hello Dr. Gaskin. i remember sometimes ago you had a video on calculating VIF using SmartPLS 3. however, now i cannot find it. Can I ask where it is? Thank you.
I show in this video how to find VIF: ruclips.net/video/J7eeu4O80_M/видео.html
great. Thank you so much.
Thank you for helping out with this video
Thanks James for the video. My model gives me a pretty decent SRMR of 0.047. However, Chi-square is infinite and NFI is n/a. Am I doing something wrong ???
Model fit is somewhat dubious in PLS still. I've found the estimates are not very consistent. So, if it is not working, I'm not surprised. I would recommend using other ways to validate the model.
@@Gaskination Thanks. Can you please elaborate other ways which can be used for model fit, if I am testing the model in SmartPLS. Thanks in advance and sorry for bothering you.
@@PraveenSrivastavaBITMesra Look at factor validity instead. Here is a video: ruclips.net/video/J_etGiwbOoM/видео.html
Hello Professor im actually writting my thesis and i got problems with my models : my sample is Low about 70 questionnaries and i have 7 variabless per model. The model fit i get (within almost 2 indicators per variable) is : SRMR = 0,049 (Estimated 0,101) and NFI = 0,86 (Estimated = 0,79). What you think about these results ? i could not reduce more the indicators.. so it's the maximum that i can obtained (I think).. please need Help :/
I would not trim your model so much to achieve good model fit in PLS. Model fit really is a foreign concept for PLS. You might also consider whether some factors are formative if you are having to trim so much.
So do I 😢
Hi James - how can I compute the adjusted chi-square? I am using SmartPLS
SmartPLS does not compute chi-square and does not rely on the covariance matrix. So, it is not provided in SmartPLS.
How do you interpret the bootstrap confidence interval and p values for SRMR? I have a case where oddly SRMR is 0.08 but the confidence interval is only between 0.01 to 0.04. When I look at the raw bootstrap values, none are even close to 0.08.
That is very odd. Luckily it is odd in the right direction. We want to minimize the SRMR, so those lower values are fine. I'm not sure though why they would be so far from the estimate...
James Gaskin thanks for the reply. Indeed. Perhaps I will play around more to see if there are some anomaly in the data contributing to this.
Hi.sir....so the saturated and estimated values for nfi, srmr has to b the same ? What can we conclude?
No. It is only has to be the same if the DF=0 (though it is possible to be the same if DF is a very small number, like less than 10)
Thank you doctor Gaskin for the valuable videos. I have a question, is it possible to get model fit indexes such as comparative fit index(CFI) and RMSEA in smartpls?
No. But you can get SRMR.
Thank you for your comment.
I AM RUNNING a model and seems that the P values almost all red and 2 factors are green ? although I have over 17 factors . What to do please
mohammed Al Yami If the p-values are not significant, that can be due to many things. It could be due to poor model fit, poor validities, poor data, poor theory, etc.
Thank you for your great video. I always suggest my students to use your vidoes for their projects and thesis writing. Do you have any academic article written using PLS SEM to use as a reference on how to report results? Thank you.
Thanks! Here is an older article: Lowry, P. B., & Gaskin, J. (2014). Partial least squares (PLS) structural equation modeling (SEM) for building and testing behavioral causal theory: When to choose it and how to use it. IEEE Transactions on Professional Communication, 57(2), 123-146.
@@Gaskination Sir you are best in my life. Thanks from China.
hi prof. please i was wandering if my model has 1 IV and 1 DV(with two dimensions) can it be possible to match the IV to the dimensions of the DV and at the same time the DV itself
I'm pretty sure SmartPLS will allow it. Although, it may be hard to interpret. I would recommend testing the model with the DV as 2nd order, and then retesting it with the DV as separate dimensions.
Dear James; Any references for these thresholds of SRMR and NFI??
Here you go: statwiki.kolobkreations.com/index.php?title=References#Model_Fit
Hi James, thanks for the excellent videos on SmartPLS, can you please let me know how to check the correlation matrix in SmartPLS? and where to find it if it is available ...? if not, what is the answer for somebody ask you about correlation matrix?
Appreciate your reply ...
Thanks
NA
You can find it in the discriminant validity section. Here is a video showing it: ruclips.net/video/J_etGiwbOoM/видео.html
When we have a variable that contains 6 items ( 3 of them are reversed ) then after analysis it still only 3 items is it ok ?
It is best to do everything you can to avoid removing items. So, I would recommend re-reversing the values of those reversed items so that they load positively on the factor. If they are still very low loadings, then I might recommend breaking this into two dimensions and then connect them to a 2nd order factor. If still no good, then yes, it is okay to lose the three.
Thank you Mr james..question please: why is the use of GOF index for PLS -SEM not advisable?
Goodness of fit is built on the assumption that there is little difference between the observed covariance matrix and the proposed covariance matrix (implied by a model). However, PLS is not a covariance-based method, so a covariance-based measure (like GOF) is fundamentally misaligned with the method.
@@Gaskination so plz what are the alternative model fit measure in PLS apart GOF
@@wafajaouadi6566 As shown in this video, they are SRMR and NFI.
@@Gaskination so can i abandon GOF index ? dont use it in my thesis
@@wafajaouadi6566 GFI is a pretty old metric that I never use, and it is not calculated in SmartPLS anyway.
This also applies for mixed (formative+reflective constructs)?
Only sort of. It's really for all reflective models. But it's really only for covariance-based SEM (not PLS). So, whatever fit measures you observe, take lightly.
Sir, how can we interpret /report partial mediation and full mediation in case of multiple mediations. thank you
If indirect path is significant, but direct paths is not, then full mediation. If indirect path is significant, and direct paths is significant, then partial mediation.
Hi James, when using smart pls, do we need to report the goodness of fit? Why?
No, not unless required by the journal and reviewers. Model fit is not particularly aligned with PLS. It is more of a maximum likelihood concept.
@@Gaskination my supervisor ask me why to compare it with Amos?
@@juanrizalsaari4635 AMOS is a more appropriate software for models that do not have formative constructs. It uses a covariance-based (maximum likelihood) approach, which is better suited to reflective models. The covariance-based approach is generally more accepted than PLS.
@@Gaskination means that when we have reflective items, are more appropriate to use PLS?
@@juanrizalsaari4635 If you have any formative factors, then PLS is best. If you only have reflective (no formative), then AMOS (or other covariance based approach) is best.
Sir, which model is checked for SRMR (saturated or estimated)? SRMR of saturated model is 0.08 but SRMR of estimated model is o.1001.... is it acceptable? or can you please give me any guideline to understand this software
Estimated. The estimated model is the one you are proposing. The saturated model estimates all possible parameters (i.e., df=0).
and what about SRMR. if SRMR is above 0.08 then what we can do. is it acceptable?
and how get the thickness of path in model which shows the significant relation
SRMR should be less than 0.800, but model fit is not totally relevant to a formative model. Joseph Hair and Christian Ringle would probaby say model fit is not the aim of PLS. As for line thickness, I'm not sure. I've never tried to play with the line formatting.
okay
Hi sir, can you please provide a video on how to check the significance (or the effect) of control variables in PLS.
same as for any predictor: just look at the p-value and regression weight.
Dear sir, how we have to make factor analysis in case of formative factors and when measured variables or indicators are having cross loading problem (in AMOS or PLS). Kindly suggest How we have to plan removal of indicators in case of formative factors if all are of significance as indicators are non interchangeable. Kindly throw some light in such type of problems. Also if possible kindly suggest the video for messy EFA specially for formative factors.
Try these two:
validating formative factors: ruclips.net/video/Qx_7pN7eAes/видео.html
factor analysis: ruclips.net/video/J_etGiwbOoM/видео.html
@@Gaskination Sir in formative factor analysis how we have to extract the factors and associated indicators ? Do we have to use same approach as that for reflective factor but in that indicators are correlated while it is not so in formative case. or will we direct have to build our own factor model then simply test the model ? But this last approach still not give outcome of maximum worth for industrial executives use or interpretation as desired by researcher. So Kindly suggest the method of extracting the factors and associated indicators in formative factors for maximum worth of the model.
Also what is the meaning of negative factor weight of indicator on formative factor as we are asking the assertive question in the survey questionnaire.
Regards
@@ArtistHridya_2023 I'm not sure I fully understand. Here is a video on how to extract factor scores (for formative and reflective): ruclips.net/video/LRND-H-hQQw/видео.html (I've queued the link to the right time point).
As for a negative factor weight, it just means that the indicator is inversely correlated with the other indicators. Make sure it is not a reverse-coded question. If it is, then re-reverse the values (so, if on 5-point Likert scale, this means subtract values from 6).
@@Gaskination Dear sir, May I have any video on messy EFA for FORMATIVE FACTORS for better model planning approach for formative model building?
Also what is the meaning of path value coefficient (in measurement model path and structural model path) if we do formative factor analysis in AMOS (as it does not seem factor weights and regression weight respectively, in AMOS as it comes in smartpls)?
@@ArtistHridya_2023 I don't have a video for a messy factor analysis with formative factors... Sorry about that. In AMOS, they are all regression weights. However, I do not recommend formative modeling in AMOS. It was not designed for this.
RMS Theta?
RMS Theta is no longer displayed.
can i retain an item with 0.6, NFI is .734
If you mean the loading is .6, then yes, that is fine.
Hello Mr. James Gaskin. I have been watching your tutorials because I am unable to get reliability of my data in PLS, if you can help me it would be very kind of you. I need help. Please.
reliability is automatically calculated for all reflective factors. Not sure what I can do to help.
James Gaskin is there any other way to contact you?
Jiya Nomi if you Google me, you will find my email.
James Gaskin Thank you so much Sir.