Not Signifincat, square root AVE is smaller than latent var
Not Signifincat, square root AVE is smaller than latent var
1. I need advice, my model is not significant, what should I do?
T Statistics (|O/STERR|)
mot -> alt 58454.596964
mot -> zak 1.089829
sat -> conf 24.403834
sat -> emp 32.314325
sat -> rel 17.296186
sat -> res 16.521170
sat -> tang 5.154653
sat -> zak 0.454472
trus -> comp 32.477463
trus -> cre 238.444918
trus -> mor 6.228259
trus -> zak 1.109754
2. Square root AVE is smaller than some of latent variable indicator in the left side, so it does not meet discriminant validity? what should I do? Thanks
alt comp conf cre emp mor mot rel res rew sat tang trus zak
alt 0.860995354
comp 0.210096 0.885650608
conf 0.260233 0.713065 0.819680426
cre 0.274107 0.830977 0.809589 0.804576907
emp 0.245635 0.611126 0.661693 0.690173 0.796508631
mor 0.269483 0.486775 0.462784 0.523366 0.407176 0.749496498
mot 0.986946 0.169225 0.218788 0.236098 0.216277 0.239133 1
rel 0.179422 0.688524 0.697875 0.709043 0.593802 0.392209 0.142318 0.874683943
res 0.092254 0.522184 0.556711 0.51711 0.663327 0.358124 0.068072 0.565079 0.943696985
rew -0.126194 0.208755 0.202115 0.178056 0.130366 0.131567 -0.284307 0.191844 0.129691 1
sat 0.247036 0.754246 0.882764 0.82096 0.880747 0.499848 0.207177 0.808293 0.789677 0.194802 0.706989392
tang 0.165972 0.453997 0.454 0.483144 0.467439 0.430381 0.139141 0.402111 0.378655 0.130859 0.579016 1
trus 0.280685 0.899506 0.810965 0.983504 0.693806 0.616723 0.23969 0.724211 0.542187 0.195071 0.833841 0.510327 0.776535254
zak 0.15158 0.136319 0.150236 0.176856 0.124038 0.149675 0.118873 0.083045 0.058275 0.170436 0.122589 -0.017373 0.178092 0.88057538
T Statistics (|O/STERR|)
mot -> alt 58454.596964
mot -> zak 1.089829
sat -> conf 24.403834
sat -> emp 32.314325
sat -> rel 17.296186
sat -> res 16.521170
sat -> tang 5.154653
sat -> zak 0.454472
trus -> comp 32.477463
trus -> cre 238.444918
trus -> mor 6.228259
trus -> zak 1.109754
2. Square root AVE is smaller than some of latent variable indicator in the left side, so it does not meet discriminant validity? what should I do? Thanks
alt comp conf cre emp mor mot rel res rew sat tang trus zak
alt 0.860995354
comp 0.210096 0.885650608
conf 0.260233 0.713065 0.819680426
cre 0.274107 0.830977 0.809589 0.804576907
emp 0.245635 0.611126 0.661693 0.690173 0.796508631
mor 0.269483 0.486775 0.462784 0.523366 0.407176 0.749496498
mot 0.986946 0.169225 0.218788 0.236098 0.216277 0.239133 1
rel 0.179422 0.688524 0.697875 0.709043 0.593802 0.392209 0.142318 0.874683943
res 0.092254 0.522184 0.556711 0.51711 0.663327 0.358124 0.068072 0.565079 0.943696985
rew -0.126194 0.208755 0.202115 0.178056 0.130366 0.131567 -0.284307 0.191844 0.129691 1
sat 0.247036 0.754246 0.882764 0.82096 0.880747 0.499848 0.207177 0.808293 0.789677 0.194802 0.706989392
tang 0.165972 0.453997 0.454 0.483144 0.467439 0.430381 0.139141 0.402111 0.378655 0.130859 0.579016 1
trus 0.280685 0.899506 0.810965 0.983504 0.693806 0.616723 0.23969 0.724211 0.542187 0.195071 0.833841 0.510327 0.776535254
zak 0.15158 0.136319 0.150236 0.176856 0.124038 0.149675 0.118873 0.083045 0.058275 0.170436 0.122589 -0.017373 0.178092 0.88057538
- Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi,
About significance
1) With bigger sample size some path will be significant, even its effect had been low. Than you should assess the “practical significance” instead the “statistical significance.
2) Keeping the current sample size.
- You could assess if the power is greater than 0.8 (I have suggested G*Power 3, it is free and you could find it at Google).
- If the Power is greater than 0.8, your sample size is enough, and you could interpret the nonsignificant path as zero (This could be sad, but zero is a result, too!).
About discriminant validity issues
The first try:
1) To have squared root of AVE greater than correlations between LV
2) The value of AVE should increase.
3) We could increase the value of AVE removing indicators with lower outer loading.
Take care with content validity, because you could delete many indicators and change the meaning of what is being measured.
The second try:
Is it possible to join the indicators of LV’s that don’t show discriminant validity (highly correlated)? Does the new LV make sense?
Best regards,
Bido
About significance
1) With bigger sample size some path will be significant, even its effect had been low. Than you should assess the “practical significance” instead the “statistical significance.
2) Keeping the current sample size.
- You could assess if the power is greater than 0.8 (I have suggested G*Power 3, it is free and you could find it at Google).
- If the Power is greater than 0.8, your sample size is enough, and you could interpret the nonsignificant path as zero (This could be sad, but zero is a result, too!).
About discriminant validity issues
The first try:
1) To have squared root of AVE greater than correlations between LV
2) The value of AVE should increase.
3) We could increase the value of AVE removing indicators with lower outer loading.
Take care with content validity, because you could delete many indicators and change the meaning of what is being measured.
The second try:
Is it possible to join the indicators of LV’s that don’t show discriminant validity (highly correlated)? Does the new LV make sense?
Best regards,
Bido
- Hengkov
- PLS Super-Expert
- Posts: 1599
- Joined: Sun Apr 24, 2011 10:13 am
- Real name and title: Hengky Latan
- Location: AMQ, Indonesia
- Contact:
Hi,
If you have second order model, you must compute AVE by hand (Repeated Indicators or Two-Stage).
Loading factor must > 0.7 for confirmatory and > 0.6 for exploratory (SQRT AVE > Correlation), If below (< 0.6) you drop it (first order and second order).
For example Statistical power using G*power (see My book SmartPLS 2.0, power = 80% or 0.8) and detail explain for analysis second-order model
Citation My books :
Latan, Hengky and Ghozali, Imam. 2012. Partial Least Squares: Konsep, Teknik dan Aplikasi SmartPLS 2.0, BP UNDIP.
Latan, Hengky and Ghozali, Imam. 2012. Partial Least Squares: Konsep dan Aplikasi Path Modeling degan XLSTAT-PLSPM, BP UNDIP (Forthcoming)
Best Regards,
Hengky
If you have second order model, you must compute AVE by hand (Repeated Indicators or Two-Stage).
Loading factor must > 0.7 for confirmatory and > 0.6 for exploratory (SQRT AVE > Correlation), If below (< 0.6) you drop it (first order and second order).
For example Statistical power using G*power (see My book SmartPLS 2.0, power = 80% or 0.8) and detail explain for analysis second-order model
Citation My books :
Latan, Hengky and Ghozali, Imam. 2012. Partial Least Squares: Konsep, Teknik dan Aplikasi SmartPLS 2.0, BP UNDIP.
Latan, Hengky and Ghozali, Imam. 2012. Partial Least Squares: Konsep dan Aplikasi Path Modeling degan XLSTAT-PLSPM, BP UNDIP (Forthcoming)
Best Regards,
Hengky
- Hengkov
- PLS Super-Expert
- Posts: 1599
- Joined: Sun Apr 24, 2011 10:13 am
- Real name and title: Hengky Latan
- Location: AMQ, Indonesia
- Contact:
Hi,
For Buy this books, please contact Prof. Imam Ghozali via email ghozali_imam@yahoo.com (Prof.Imam sent for you from Semarang).
Please Used Two-Stage Approach for second order model after repeated indicators approach and check collinearity problem using WarpPLS. We have books WarpPLS too.
Regards,
Hengky
For Buy this books, please contact Prof. Imam Ghozali via email ghozali_imam@yahoo.com (Prof.Imam sent for you from Semarang).
Please Used Two-Stage Approach for second order model after repeated indicators approach and check collinearity problem using WarpPLS. We have books WarpPLS too.
Regards,
Hengky
- Hengkov
- PLS Super-Expert
- Posts: 1599
- Joined: Sun Apr 24, 2011 10:13 am
- Real name and title: Hengky Latan
- Location: AMQ, Indonesia
- Contact:
Hi Dodik,
In my book XLSTAT-PLSPM (fortcoming) with Prof. Imam, we explain this procedure (Two-Stage Approach for analysis second-order model).
Stage 1: you analysis orginal model (for second-order construct used repeated indicators), evaluation outer model and save latent score.
Stage 2: Input latent score with mode B and evaluation inner model.
Good luck.
Best Regards,
Hengky
In my book XLSTAT-PLSPM (fortcoming) with Prof. Imam, we explain this procedure (Two-Stage Approach for analysis second-order model).
Stage 1: you analysis orginal model (for second-order construct used repeated indicators), evaluation outer model and save latent score.
Stage 2: Input latent score with mode B and evaluation inner model.
Good luck.
Best Regards,
Hengky
- Hengkov
- PLS Super-Expert
- Posts: 1599
- Joined: Sun Apr 24, 2011 10:13 am
- Real name and title: Hengky Latan
- Location: AMQ, Indonesia
- Contact:
Hi Dodik,
In my book XLSTAT-PLSPM (forthcoming) with Prof. Imam, we explain this procedure (Two-Stage Approach for analysis second-order model).
Stage 1: you analysis orginal model (for second-order construct used repeated indicators), evaluation outer model and save latent score.
Stage 2: Input latent score with mode B and evaluation inner model.
Good luck.
Best Regards,
Hengky
In my book XLSTAT-PLSPM (forthcoming) with Prof. Imam, we explain this procedure (Two-Stage Approach for analysis second-order model).
Stage 1: you analysis orginal model (for second-order construct used repeated indicators), evaluation outer model and save latent score.
Stage 2: Input latent score with mode B and evaluation inner model.
Good luck.
Best Regards,
Hengky
- Hengkov
- PLS Super-Expert
- Posts: 1599
- Joined: Sun Apr 24, 2011 10:13 am
- Real name and title: Hengky Latan
- Location: AMQ, Indonesia
- Contact:
Hi Dodik,
In my book XLSTAT-PLSPM (see PLS Literature books-forthcoming), we explain this procedure (Two-Stage Approach for analysis second-order model).
Stage 1: you analysis orginal model (for second-order construct used repeated indicators), evaluation outer model and save latent score.
Stage 2: Input latent score with mode B and evaluation inner model.
Good luck.
Best Regards,
Hengky
In my book XLSTAT-PLSPM (see PLS Literature books-forthcoming), we explain this procedure (Two-Stage Approach for analysis second-order model).
Stage 1: you analysis orginal model (for second-order construct used repeated indicators), evaluation outer model and save latent score.
Stage 2: Input latent score with mode B and evaluation inner model.
Good luck.
Best Regards,
Hengky
Thanks bro, I have do it, but it is not significant.Hengkov wrote:Hi Dodik,
In my book XLSTAT-PLSPM (forthcoming) with Prof. Imam, we explain this procedure (Two-Stage Approach for analysis second-order model).
Stage 1: you analysis orginal model (for second-order construct used repeated indicators), evaluation outer model and save latent score.
Stage 2: Input latent score with mode B and evaluation inner model.
Good luck.
Best Regards,
Hengky