Discriminant validity

This forum is closed, and read-only.
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Hi Mahmood,
Rule of Thumb evaluation outer model:
AVE > 0.50
Communality > 0.50
Cronbach's Alpha > 0.70
Composite Reliability > 0.70
For your model is not valid.
Delete indicators your constrcut loading factor < 0.70 = AVE > 0.50
Regards,
Hengky
Mahmood
PLS Expert User
Posts: 58
Joined: Sun Dec 18, 2011 11:08 am
Real name and title:

Post by Mahmood »

Thanks for your comments.
I did and I performed the new analysis. However, one of my latent variables got the value of AVE and Cronbachs Alpha 1.000000? This item had two questions in questionnaire, which I deleted one of them as you said. Is this Confirmatory factor analysis?
Do I need to report the deleted items?
Last but not least, two other constructs Cronbachs decreased to below .70 ! Do I need to remove them? But these are new constructs and they are important to keep!!!!

This is a new model that I am proposing, but the questions are based on different valid questionnaires.
Mahmood
vonbergh
PLS Expert User
Posts: 60
Joined: Tue Jan 26, 2010 10:06 pm
Real name and title:

Post by vonbergh »

Hi Mahmood,

With a Cronbach's alpha of ,27 the reliability of the Variable (IV or DV) is too low. This would have been the case in SmartPLS and in SPSS, so you'll have to look at how the questions in the questionnaire predict the variable. Were they translated correctly, are they clear enough, do they measure the same variable etc etc. Also multivariate approaches (e.g. factor analysis) may be helpful.

Good luck,
Dennis
mzhang
PLS Junior User
Posts: 1
Joined: Mon Oct 11, 2010 2:36 pm
Real name and title:

About AVE

Post by mzhang »

What is one of my AVE is about equal to one of the correlation?
Is this ok?
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Hi, Mahmood.
Your result AVE and Cronbach's Alpha is 1.000? You questionnaire used likert scale?
For your construct some literature suggest used > 3 items- 6 items measures.
Cronbach's Alpha PLS under estimate used Dillon Goldenstein.
For report please see Chin, W.W. (2010) How to Write up PLS Analysis. Handbook PLS, Berlin-Spinger.

Hi, Zhang.
Rule of Thumb square root average variance extracted > correlation.
For your case equal => acceptable, if cross loadings > 0.70 and indicator's loadings should be higher than all of its cross loadings.
Regards,
Hengky.
vonbergh
PLS Expert User
Posts: 60
Joined: Tue Jan 26, 2010 10:06 pm
Real name and title:

Re: About AVE

Post by vonbergh »

Hi Zhang,
mzhang wrote:What is one of my AVE is about equal to one of the correlation?
Is this ok?
In addition to Hengky's post, I would investigate your data for multicollinearity. See a.o. Field, A.P. (2010) Discovering Statistics Using SPSS, Los Angeles, CA: Sage.

Good luck,
Dennis
Mahmood
PLS Expert User
Posts: 58
Joined: Sun Dec 18, 2011 11:08 am
Real name and title:

Post by Mahmood »

Hengkov wrote:Hi, Mahmood.
Your result AVE and Cronbach's Alpha is 1.000? You questionnaire used likert scale?
For your construct some literature suggest used > 3 items- 6 items measures.
Cronbach's Alpha PLS under estimate used Dillon Goldenstein.
For report please see Chin, W.W. (2010) How to Write up PLS Analysis. Handbook PLS, Berlin-Spinger.

Hi, Zhang.
Rule of Thumb square root average variance extracted > correlation.
For your case equal => acceptable, if cross loadings > 0.70 and indicator's loadings should be higher than all of its cross loadings.
Regards,
Hengky.
I dropped the items with less than 0.70 ( outer loading) and the Cronbachs Alpha as well as AWE of this construct changed to 1.00000. why it is 1?
This is the construct that has the Cronbachs 0.22 before I drop that items.

Thanks for your help.
Mahmood
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Hi Mahmood,
Your construct in model reflective or formative?
If formative, you not assess validity and reliability construct (loading factor, AVE, Cronbach Alpha etc), but you look significant weight.

For you result AVE = 1, please check problem collonierity.
You items used likert scale, dichotomus, or?
Cronbach Alpha PLS under estimate, look composite reliability (if reflective model).
Best Regards,
Hengky
Mahmood
PLS Expert User
Posts: 58
Joined: Sun Dec 18, 2011 11:08 am
Real name and title:

Post by Mahmood »

Hengkov wrote:Hi Mahmood,
Your construct in model reflective or formative?
If formative, you not assess validity and reliability construct (loading factor, AVE, Cronbach Alpha etc), but you look significant weight.

For you result AVE = 1, please check problem collonierity.
You items used likert scale, dichotomus, or?
Cronbach Alpha PLS under estimate, look composite reliability (if reflective model).
Best Regards,
Hengky
Thanks for your help.
May I ask you to have a look at the overall quality of this study? I added the image of boots raping and PLS here in this link:
http://www.hajili.com/uncategorised/pls-algorithm
Mahmood
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Hi Mahmood,
I'm look your result model, you drop only indicator FC1 (Social Commerce Components) => model fit. For another construct with loading factor > 0.50 acceptable (see Chin 1998), because your indicator construct only 2, 3 and 4 (can't delete). Recomendation some literature for measures construct (indicators 5-6 items).
Chin, W. W. 1998. The partial least squares approach for structural equation modeling. in G. A. Marcoulides (Ed.), Modern methods for business research (pp. 295-236). London: Lawrence Erlbaum Associates.
Regards,
Hengky
Mahmood
PLS Expert User
Posts: 58
Joined: Sun Dec 18, 2011 11:08 am
Real name and title:

Post by Mahmood »

Hengkov wrote:Hi Mahmood,
I'm look your result model, you drop only indicator FC1 (Social Commerce Components) => model fit. For another construct with loading factor > 0.50 acceptable (see Chin 1998), because your indicator construct only 2, 3 and 4 (can't delete). Recomendation some literature for measures construct (indicators 5-6 items).
Chin, W. W. 1998. The partial least squares approach for structural equation modeling. in G. A. Marcoulides (Ed.), Modern methods for business research (pp. 295-236). London: Lawrence Erlbaum Associates.
Regards,
Hengky
Thanks for your comment.
Shall I drop FC1 and do PLS algorithm and Bootsraping again and show the new results in my paper?
Mahmood
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Hi Mahmood,
Yes.
For item delete, you report used footnote.
Regards,
Hengky
Mahmood
PLS Expert User
Posts: 58
Joined: Sun Dec 18, 2011 11:08 am
Real name and title:

Post by Mahmood »

Hengkov wrote:Hi Mahmood,
Yes.
For item delete, you report used footnote.
Regards,
Hengky
thanks.
Mahmood
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Ok Mahmood,
Your welcome and good luck.
Regards,
Hengky
Mahmood
PLS Expert User
Posts: 58
Joined: Sun Dec 18, 2011 11:08 am
Real name and title:

Post by Mahmood »

Hengkov wrote:Hi Mahmood,
Your construct in model reflective or formative?
If formative, you not assess validity and reliability construct (loading factor, AVE, Cronbach Alpha etc), but you look significant weight.

For you result AVE = 1, please check problem collonierity.
You items used likert scale, dichotomus, or?
Cronbach Alpha PLS under estimate, look composite reliability (if reflective model).
Best Regards,
Hengky
Thanks again.
So, for the dependent variable or formative construct, I don’t need the questionnaire items. Do I?
As I read the papers I don’t see they measure the formative construct .
If not how can I see the R square on that?
I am designing a new model now.
Mahmood
Locked