SmartPLS 4: Testing structural hypotheses

[@arifashraf8569]

Thank you Prof. Gaskin for always guiding us

[@Gaskination]

I had never heard of CVPAT until your comment :)

[@arifashraf8569]

@@Gaskination Could you please make one :) I would love to include in my latest research paper

[@Gaskination]

@@arifashraf8569 I looked into it yesterday. I'll probably be making another few videos next week.

[@arifashraf8569]

@@Gaskination Thank you Prof. Gaskin

[@reanalytics1863]

I got the feeling exporting just got more time consuming with SmartPLS 4 However, It shows that it has promising new features that will streamline the overall workflow and improve of the quality of results

[@DorinaNicoaraPopescu]

Hello again James. Another questions: What is the difference between "standard algorithms" and "consistent algorithms". I am getting better results with a stardard version........ many thanks

[@Gaskination]

There are two sections of algorithms. You'll see the section for consistent in the menu. It is for models with all reflective factors. However, the consistent algorithm often creates odd anomalies. It is supposed to correct for possible inflation of estimates when all factors are reflective.

[@MsEndOfReason]

Thank you so much for those tutorials. They are extremely helpful. One question that I could not find the answer for so far is what is the difference between the path and the factor analysis? And when do I use which type?

[@Gaskination]

Path weighting is the default and is generally recommended for testing hypotheses. It also results in the highest R2 values. Factor weighting is appropriate when assessing just the measurement model. Despite these differences, the estimates produced will only differ very slightly, if at all.

[@MsEndOfReason]

@@Gaskination Thank you so much!

[@TUPhDude]

Hi Dr. Gaskin As always, thanks for the video. I am wanting to know your opinion on when a coefficient value is too high and how to deal with that? I have a coefficient value of just over 1 and I am concerned that it will be a problem. I have checked for multicollinearity and all VIF are well below 3.3. Any advice?

[@Gaskination]

unstandardized coefficients have no restrictions (except based on the original scales of the variables involved). If a standardized coefficient is greater than 1.00, then this implies some error in the calculation, usually due to including categorical variables, or due to extreme non-normality, or low sample size.

[@farahfarhana4014]

Prof., what if I got high R square like 0.83, Multicollinearity OK but read that it can also shows spurious causation. How do we test for the spurious causation? Or any explanation that can justify the high R Square? It is exploratory research (extension of UTAUT), 7/11 relationships are significant.

[@Gaskination]

UTAUT has notoriously high correlations among factors. As long as you have ruled out discriminant validity (and maybe method bias), then the high R-square can be attributed to strong measures for highly correlated constructs.

[@مسلمعبدالله-س7ن]

I have a question please. In a simple mediation model and after conducting PCA on SPSS, which is better, to continue with SPSS and test hypotheses using regression, or to test the hypotheses using PLS-SEM on SmartPls, given that the two approaches give different results.

[@Gaskination]

SPSS can handle simple regressions, whereas SmartPLS can handle SEM, including complex models that involve mediation and moderation.

[@sowmyakini8730]

Thank you for your helpful videos, Prof. Gaskin. However, with a complex model, on running a consistent PLS bootstrapping, all the P values of path coefficients are coming to Zero. Does the condition "consistent pls bootstrap needs to be used if all latent factors are reflective in nature" hold true? Running just bootstrapping gives mixed p values. Kindly let me know why this difference is observed.

[@Gaskination]

Sometimes PLSc seems to create erroneous results. If your results look unreasonable, revert to the regular algorithms.

[@findyourway-withnayabkaukab]

Thank you Professor for this wonderful video! However, is it necessary to run PLSc algorithm? Is there a reference that supports to use standard algorithm even though constructs are reflective?

[@Gaskination]

Now that CB-SEM is available in SmartPLS, I would recommend not using PLSc if all factors are reflective. Instead just use CB-SEM.

[@findyourway-withnayabkaukab]

@@Gaskination Thank you so much for your response. My proposal was based on PLS SEM so I'm afraid I can't switch to CB-SEM. Is there any reference that supports standard algorithm even though all but one constructs are reflective?

[@Gaskination]

@@findyourway-withnayabkaukab If one construct is formative then you wouldn't use CB-SEM anyway. The Lowry & Gaskin PLS article states this. If you have any formative factors, then PLS is a more appropriate approach than CB-SEM. 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.

[@findyourway-withnayabkaukab]

@@Gaskination Hi Prof., One more question please :) If all the constructs are reflective, is there any source that I can cite to justify the selection of using standard PLS SEM? The problem is that I am not getting the desired results from PLSc

[@Gaskination]

@@findyourway-withnayabkaukab I'm not sure of one in particular. But there are many "PLS is awesome" papers listed here that might help: www.smartpls.com/documentation/

[@Albert-i6j]

Thanks for your wonderful mini-lectures! I wonder where to put a covariate LV for the DV in a similar model like in your video?

[@Gaskination]

You can add it just like you might add any predictor. In the video it would be something like the way I modeled ethical concerns.

[@DorinaNicoaraPopescu]

Hello James. One question: when p value is = 0.000, but the path coefficients are below 0.20? can I confirm the relationships between constructs?

[@Gaskination]

Yes, but I would recommend looking at the effect size (f-square), especially if you have a large sample size. The path coefficients can also be very small if not standardized (depending on the scale).

[@مسلمعبدالله-س7ن]

Hello Dr. and thanks for the video. Where can I finde the R square that is caused by each variable not the collective R square caused by all exogenous variables?

[@Gaskination]

It is not listed. You would have to check the correlation matrix and then square the r value. Or you could remove the variable of interest to see the drop in R-square.

[@مسلمعبدالله-س7ن]

​@@GaskinationThanks a lot Dr

[@bongiwesithole2145]

Thank you for the video. I have two questions, 1. is it an issue that the means and SD's of my 5000 samples are zero? 2. my model has independent variables (IV) which are predicting intention, intention is then predicting usage. I get N/A R2 values for the IV to intention relationship when using consistent pls bootstrapping, what am I doing wrong?

[@Gaskination]

If mean AND standard deviation are zero, then that implies there is zero variance. In this case, the variable does not vary. So, it is useless. If the mean is zero and the standard deviation is 1.00, then that means it is standardized. The PLSc error is common. It just means it couldn't calculate it. In this case, switch to regular PLS algorithm.

[@bongiwesithole2145]

@@Gaskination thank you!

[@sihirbazjack]

Hello Prof. Gaskin, my hypothesis has p value exactly 0.05 in smartpls 4. But it has 0 value in confidence intervals bias corrected. Can I report that the hypothesis is supported?

[@Gaskination]

0.05 is just a suggested target. It’s not a strict cut off. I’d say you’re good to go.

[@tombrill-g8y]

Most of your video's utilize a two-tail test type. When should we use a one-tailed approach vs. a two-tailed approach?

[@Gaskination]

In practice, most just use two-tailed all the time. However, in theory, if you have already theorized about the direction of an effect, or if the direction of the effect is known a-priori, then a one-tailed test is fine.

[@tombrill-g8y]

Thank you! @@Gaskination

[@hameemislam753]

@ James Gaskin Sir i have performed second order cfa and now checking the structural model i am confused about the relationship of Acr on ENG whether itz significant or not becoz its beta is 0.090 sample mean is 0.091 std dev is 0.029 t- stat is 3.143 and p value is 0.001 and itz in green sir is this hypotheses is significant or insignificant becoz its path coff is less

[@Gaskination]

I would check the effect size. If the effect size is at least small (not "no effect"), then I would claim the path is not nothing (you can reject null hypothesis).

[@mahamadoukante9463]

Thanks for your efforts Prof. I am having weird results using the PLSc algorithm on an all-reflective model. My R square is higher with PLSc but some of the loadings are greater than one and many became lesser than .7. With the normal algorithm, even though the R square is smaller, all items load well with good AVEs. What should I do?

[@Gaskination]

I often struggle with the PLSc algorithm as well. The results seem ironically inconsistent. For this reason, when I have a fully reflective model, I use AMOS or Mplus.

[@mahamadoukante9463]

@@Gaskination Understood!

[@kasoomro]

What are the recommended values for a significant hypothesis?

[@Gaskination]

Usually we aim for p-value < 0.05 (i.e., 95% confidence), but there are reasons to aim higher or lower, depending on context.

[@DorinaNicoaraPopescu]

Hello again. How can I add a "control variable" for instance "company's sector" or "company's size" to an existent model with 4 reflective factors ? what are the steps? Thank you

[@Gaskination]

If it is something continuous or ordinal, like company size, then you can add it as an indicator to a factor that is predicting all endogenous factors. If it is something nominal/categorical like company sector, then you would break it into a set of dummy (binary) variables each representing one sector (except the reference category), and then include each dummy variable on a different factor, each as a predictor of all endogenous factor.

[@hediek.m6279]

Thank you so much for these tutorials. One issue that I can not find a solution to is how to use binary predictive indicators. Should I make them all Metric?

[@Gaskination]

There is a binary option now in the data setup. Here is a video that shows it: kzbin.info/www/bejne/mqa9kKBjfMSborc

[@DorinaNicoaraPopescu]

Hello. one question: when running bootstraping for structural model, I am getting P-values NaN. what can I do? thanks

[@Gaskination]

If you are using consistent algorithms, try switching to regular bootstrapping. If not, try running without bootstrapping to see if you get odd values.

[@DorinaNicoaraPopescu]

Thanks a lot for your answer @@Gaskination ! . Another questions: What is the difference between "standard algorithms" and "consistent algorithms". I am getting a p value only if I use "stardard version of bootstraping" not the consistent........ is that ok? many thanks

[@Gaskination]

@@DorinaNicoaraPopescu There are two sections of algorithms. You'll see the section for consistent in the menu. It is for models with all reflective factors. However, the consistent algorithm often creates odd anomalies. It is supposed to correct for possible inflation of estimates when all factors are reflective.

[@clivechifamba179]

Hi. I am having an issue. when I remove the insignificant paths and rerun the analysis some factor loadings are changing. what could be the cause of this?

[@Gaskination]

You do not have to remove non-significant paths. The reason they change is because SEM is a network of interrelated and interdependent effects. So, when we remove or add any parameter, it will change everything else.

[@amritakulshreshtha2075]

I am getting n/a values in sample mean after running consistent pls algorithm with bootstrapping. How to deal with that issue.

[@Gaskination]

This is still a common problem that has not been fixed. Unfortunately, the best thing you can do is run the regular PLS algorithm

[@divinelumiere8564]

Please Prof, What about the blindfolding procedure? I don't see it in the "calculate" button

[@Gaskination]

It is now consolidated into its respective relevant other analyses. It is no longer a separate option.

[@divinelumiere8564]

@@Gaskination Thank you 🙏🙏

[@aliciaregina7679]

hello professor, may I know how to find beta coefficients in SmartPLS 4?

[@Gaskination]

Beta is the regression weight. So, it's just listed in the direct effects in the report.

[@atharmukhtar7326]

Could you tell how i can find estemation value??

[@Gaskination]

The estimates are just the coefficients. So, in the case of the structural model, that would be the regression weight or beta.

[@weamelnaghy9981]

Please professor, what about a-square ? How could I found it in smartpls4

[@Gaskination]

I've never seen a-square on smartPLS 4.

[@MS-mx5of]

may i know is it logical if i have many p-values = 0.000 ?

[@Gaskination]

This is just the way SmartPLS shows p-values less than 0.001 (rather than giving the exact p-value down to more than three decimals). So, yes, this is totally fine. When reporting it, I would recommend to report it as less than 0.001, rather than equal to 0.000

[@enasshiha8021]

what are the range of accepted values of p values and T values

[@Gaskination]

easy to google. t values indicate p-values less than 0.05 when they are above 1.96 (two tailed). p-values less than 0.05 are conventional for "significance", but really it depends on many things.

[@data-sushi]

Please Prof, How to calculate the mediating effect

[@Gaskination]

See my two most recent videos here: kzbin.info They are both about mediation in SmartPLS 4

[@antonioliriololi]

How to get that beta version?

[@Gaskination]

The beta version is not available to the public. The public version should be released within the next couple of weeks. Hopefully early June.

[@antonioliriololi]

@@Gaskination Thank you

[@kevinwilliam5496]

why my smartpls 4 doenst have report button after calculate

[@Gaskination]

There's just a little checkbox for it in the calculation window. If you don't check it, then look on the bottom left after the model runs and you'll see a button to open report.

[@kevinwilliam5496]

@@Gaskination oh yeah you right thank you so much now i can continue my thesis

[@hanspirin]

Thanks for the video