Dr, I have a question about running PLS-predict. Since both endogenous variables in my model are second-order constructs. Is it reasonable to report Q² values, MAE, and RMSE at the sub-dimension level rather than at the indicators level?
Thanks for the video! One indicator in one of my constructs has a Q2 < 0. Other than that, the model has very good predictive power. This indicator has an outerloadings of only 0.59 too, so I was wondering if I should delete this indicator (after deleting, everything seems perfect). However, this indicator plays an important aspect in the scale and this scale is also from a previous tested research. It is theoretically not a good idea to remove it, and the AVE of the construct is above 0.5 too with this indicator. Would you recommend I remove it? Or if I do not remove it, how can I explain Q2
You could report both. Usually in science, it is best to disclose information. So, you might add an appendix or footnote that shows the alternative approach.
Prof, I have a question about PLSpredict/CVPAT. when I start calculating, my data have singular matrix problem, but when I start calculating algorithm or bootsrap, it works. how can I solve this problem, Prof?
Usually this means there is a group in one of the predictors that is related to a value in the DV with zero variance. For example, if one of the predictors is autonomy and the DV is job satisfaction, perhaps anyone who responded with 5/5 for autonomy also responded 5/5 for job satisfaction, without any exception (variance). I would recommend omitting one predictor at a time (probably one that doesn't directly predict the DV in the regular model) until you find the variable causing the issue.
Wow, smart-pls has finally started reached to actual predictive modeling. Wonder how this will change the upcoming research in a few years where top quality journals like isr may start asking for this.... Whether it makes sense to do it or not.
That would require multiple linear regression, or you can input the unstandardized regression weights into an MLR formula, including the intercept. I don't have a video for that though.
Hello Prof. Can you help explain how we assess and interpret the results of PLS Predict with higher order dependent variable? Do we use latent score for this analysis? Do share the link in case of any video already made on it. Thanks
It should work the same as in this video unless the higher order dependent variable is formatively measured (arrows point from first order dimensions to 2nd order DV). If the dependent variable is higher order formative, then yes, you'll need to use latent variable scores, as shown in this video: ua-cam.com/video/LIz9QQ-3g6I/v-deo.html
Hi james, thank you so much for this useful information. I want to request if you can please give more insight on difference between pls blindfolding in smartpls3 and pls predict in pls4.. and also i am getting problem in finding about q squsre effect size. Where can we find it in pls4?
@@ingyselim2760 hi.m we have pls predict in pls4. Blindfolding procedure has been discontinued as it takes only in sample data, whereas pls predict takes bothe in sample and out sample data
@@nehavarma8654 and how to evaluate the Q square effect size?? Shall I conduct it manually similar to SmartPLS 3? Q square effect size= (q2 included-q2 excluded)/(1-q2 included)
@@ingyselim2760 as per my knowledge pls predict is showing q square effect soze only as it is calculated on included and exculded both.. though m not sure as much resource isnt available regarding this. So i think expert can solve this.
Hello Prof Thank you for all the resources you have been sharing I have a worry, I was unable to locate the BLINDFOLDIING procedure in SmartPLS 4 Has it been changed? or eliminated?
You can add them, but the PLS Predict algorithm does not create any output for them. To see the effect for interactions, run the regular PLS algorithm (and bootstrap).
Thank you Prof. Gaskin for your consistent scholarly presentation and support in Smart-PLS. I really admire your intellectual prowess. God bless you
Simple and insightful, thank you so much
Dr, I have a question about running PLS-predict.
Since both endogenous variables in my model are second-order constructs. Is it reasonable to report Q² values, MAE, and RMSE at the sub-dimension level rather than at the indicators level?
Yes
@@Gaskination Thank you very much
hi Prof, if we use PLS Predict, we should kick the structural model discussion?
I don’t think I understand your question
Thanks for the video! One indicator in one of my constructs has a Q2 < 0. Other than that, the model has very good predictive power. This indicator has an outerloadings of only 0.59 too, so I was wondering if I should delete this indicator (after deleting, everything seems perfect). However, this indicator plays an important aspect in the scale and this scale is also from a previous tested research. It is theoretically not a good idea to remove it, and the AVE of the construct is above 0.5 too with this indicator. Would you recommend I remove it? Or if I do not remove it, how can I explain Q2
You could report both. Usually in science, it is best to disclose information. So, you might add an appendix or footnote that shows the alternative approach.
@@Gaskination Great idea! Thanks, professor! Have a great day!
Thank you Prof. (Dr.) Gaskin.
In my research out of 6 constructs, one is negative (Q2), remaining five are above 0 (Q2). How to interpret this?
It just means that the model is not good for predicting that variable. But the rest is fine.
Thank you professor for your reply 🙏 Just to inform that blindfolding is providing all six constructs above 0. So, I am planning to report both.
Prof, I have a question about PLSpredict/CVPAT. when I start calculating, my data have singular matrix problem, but when I start calculating algorithm or bootsrap, it works. how can I solve this problem, Prof?
Usually this means there is a group in one of the predictors that is related to a value in the DV with zero variance. For example, if one of the predictors is autonomy and the DV is job satisfaction, perhaps anyone who responded with 5/5 for autonomy also responded 5/5 for job satisfaction, without any exception (variance). I would recommend omitting one predictor at a time (probably one that doesn't directly predict the DV in the regular model) until you find the variable causing the issue.
Prof. Shall I do this individually for all the groups in my data or just the overall data set?, as I have to do mga later
Yes, if you want to understand the predictive quality of the model for each group.
Wow, smart-pls has finally started reached to actual predictive modeling. Wonder how this will change the upcoming research in a few years where top quality journals like isr may start asking for this.... Whether it makes sense to do it or not.
what if any of the groups has negative q square values.
then it doesn't have good predictive quality
When given new data, how do you predict new values using the model?
That would require multiple linear regression, or you can input the unstandardized regression weights into an MLR formula, including the intercept. I don't have a video for that though.
Shall I do this for all the groups in my data, as I have to do mga
Yes, if you want to understand the predictive quality of the model for each group.
Is Smart pls 4 able to process data using pls predict cvpat?
I'm not sure. I did not come across this feature.
Hello Prof. Can you help explain how we assess and interpret the results of PLS Predict with higher order dependent variable? Do we use latent score for this analysis? Do share the link in case of any video already made on it. Thanks
It should work the same as in this video unless the higher order dependent variable is formatively measured (arrows point from first order dimensions to 2nd order DV). If the dependent variable is higher order formative, then yes, you'll need to use latent variable scores, as shown in this video: ua-cam.com/video/LIz9QQ-3g6I/v-deo.html
Thank you prof.
Hi james, thank you so much for this useful information. I want to request if you can please give more insight on difference between pls blindfolding in smartpls3 and pls predict in pls4.. and also i am getting problem in finding about q squsre effect size. Where can we find it in pls4?
Here is a good resource for this question: smartpls.com/documentation/algorithms-and-techniques/predict
Hello Ms Neha,
Did you find the Q square effect size on smartPLS 4?
@@ingyselim2760 hi.m we have pls predict in pls4. Blindfolding procedure has been discontinued as it takes only in sample data, whereas pls predict takes bothe in sample and out sample data
@@nehavarma8654 and how to evaluate the Q square effect size?? Shall I conduct it manually similar to SmartPLS 3?
Q square effect size= (q2 included-q2 excluded)/(1-q2 included)
@@ingyselim2760 as per my knowledge pls predict is showing q square effect soze only as it is calculated on included and exculded both.. though m not sure as much resource isnt available regarding this. So i think expert can solve this.
Hello Prof
Thank you for all the resources you have been sharing
I have a worry, I was unable to locate the BLINDFOLDIING procedure in SmartPLS 4
Has it been changed? or eliminated?
The blindfolding procedure is now built into other procedures and is no longer available on its own.
Thanks ..I have negative value how can I explain it?
regression weights can be negative; it just means that as one variable goes up, the other goes down
@@GaskinationI have smartpls 3 .I calculate Q square predictive manual. I have negative value
@@lakhdarchibout8826 Oh, for the Q2. That means you don't have a good predictive value. That means your model is not good at predicting the outcome.
Hi, can I add interaction term for pls predict
You can add them, but the PLS Predict algorithm does not create any output for them. To see the effect for interactions, run the regular PLS algorithm (and bootstrap).
Thank you...