GoF threshold

[chinasun]

hi all.
a goodness-of-fit of gloabl model occured in Tenenhaus et al paper(2005), proposed by Amato et al ,but it didnt show the threshold of GoF. Who can tell me the threshold of gof?
thanks in advance!

[Diogenes]

Hi Sun,
There is not a threshold, but if we think that the AVE should be > 0,5
and that the GoF is like a geometrical mean of the measurement and structural model. We could say that a GoF > 0,5 also will be Ok (my opinion, without references).

Best regards.
Bido

[Diogenes]

See viewtopic.php?t=1094&highlight=gof

Reference and threshold based in effect size.

Best regards,

Bido

[ruchi]

Does smartPLS gives this GoF. ? if it does not then how can i calculate it for my model?

Thanks
ruchi

[Diogenes]

Hi,
you should compute it by hand

see
TENENHAUS, M.; VINZI, V. E.; CHATELIN, Y.; LAURO, C. PLS path modeling. Computational Statistics & Data Analysis, v.48, p.159-205, 2005.
https://studies2.hec.fr/jahia/webdav/si ... S_PM_5.pdf


Best regards,


Bido

[ruchi]

Thanks a lot,
I will read it.

[ruchi]

Hello

I tried to read the paper and got some concepts cleared regarding GoF. will it applicable if i have some of the formative indicators?

Thanks
Ruchi

[ruchi]

Hello

I calculated the GoF and it came out to be 0.26. Can someone tell me if this GoF is ok and what does it indicate.

Thanks
Ruchi

[Diogenes]

Hi,

See p.187:
WETZELS, M. et al. Using PLS path modeling for assessing hierarchical construct models: guidelines and empirical illustration. MIS Quarterly, v.33, n.1, p.177-195, March 2009.

AVE ≥ 0,5
R2 from Cohen: small: 0.02; medium: 0.13; large: 0.26


Classification of GoF, from Wetzels et al.
GoFsmall=0.1, GoFmedium=0.25, and GoFlarge=0.36


Best regards,

Bido

[ruchi]

Dear Sir

Thanks for the reply
I was trying to locate the paper, but could not access it or download it.
Can u please provide me this article (WETZELS, M. et al. Using PLS path modeling for assessing hierarchical construct models: guidelines and empirical illustration. MIS Quarterly, v.33, n.1, p.177-195, March 2009. ) so that i can look into this.

In smartPLS- I m getting R square- 0.335( represented in blue circle of my dependent variable, measured by 4 items). Now when i remove my one item (D4) because in intiial factor analysis in spss its loading was less than 0.5,my Rsquare drastically gets reduced to 0.214 .

If i add this item and do PLS then its loading is more than 0.5 and R sqaure also is 0.335.

I dont know why its happening.
I think Ideally i should not keep this item as in the initial analysis its factor loading was less than o.5 and it should be removed from PLS analysis.

It will be great if you can please help me in resolving this confusion.

Thanks
Ruchi

[Diogenes]

Hi Ruchi,

- When we run a factor analysis in SPSS, the estimation does not consider that the factor is correlated with others. PLS-PM does this, and in your model it seems that this “excluded” indicator is highly correlated with the other LV.
- For this reason, currently I have used the two step approach that is common in LISREL, first I run a confirmatory factor analysis (connect all LV – one with another – use the factor weighting scheme, assess the outer loadings, AVE, reliability, correlation between LV). In the second step, run the structural model.

Best regards,

Bido

[ruchi]

Thanks

It means i should not compare with factor analysis in spss (using varimax rotation and Principal component) with factor analysis is PLS. After checking for cronbach alpha of my constructs , I should make my model (connecting all LV) and run PLS with factor loading weighting scheme .

After checking for reliability and validity , i should run my model using PLS with Path weighting scheme with bootstrapping.

Please correct me if i have not understood it .

Thanks
Ruchi