Dear Wali,

Testing multivariate normality is a crucial step if one is using co-variance based technique (AMOS), whereas its not a requirement for Smart PLS which is non-parametric technique. Smart PLS does not do any assumption regarding the distribution of the data whereas AMOS is very sensitive to this. In order to check the multivariate normality simple follow these steps in AMOS:

-Go to Analysis Properties

- Go to Output section

- Check "Test for normality and outliers"

- Close the box and

- Then "Calculate Estimates"

Check the Critical ration (c.r.) which has to be below the acceptable range of < 5 (Bentler, 2005).

Byrne (2010) recommends following steps to rectify the multivariate normality issues.

i)  Assessment of multivariate outliers

ii) To choose between Asymptotic distribution free (ADF) estimation, or Satorra and Bentler (1988, 1994) estimation or Bootstrapping techniques.

If the problem still persists then you would need to run Bootstapping technique in the last resort. Byrne (2010) considers bootstrapping technique superior in such cases. 

Best of luck

Dear Wali,

Testing multivariate normality is a crucial step if one is using co-variance based technique (AMOS), whereas its not a requirement for Smart PLS which is non-parametric technique. Smart PLS does not do any assumption regarding the distribution of the data whereas AMOS is very sensitive to this. In order to check the multivariate normality simple follow these steps in AMOS:

-Go to Analysis Properties

- Go to Output section

- Check "Test for normality and outliers"

- Close the box and

- Then "Calculate Estimates"

Check the Critical ration (c.r.) which has to be below the acceptable range of < 5 (Bentler, 2005).

Byrne (2010) recommends following steps to rectify the multivariate normality issues.

i)  Assessment of multivariate outliers

ii) To choose between Asymptotic distribution free (ADF) estimation, or Satorra and Bentler (1988, 1994) estimation or Bootstrapping techniques.

If the problem still persists then you would need to run Bootstapping technique in the last resort. Byrne (2010) considers bootstrapping technique superior in such cases. 

Best of luck

Thank you sir. Need you support for this. As ADF requires a large data set probably more than 1000 observations. Satorra and Bentler estimation are not available in the AMOS. Can you specifically guide me in the context of the AMOS, Can I use GLS (Weighted Least Square) option, which is readily available in the AMOS?

Can you guide me through step by step process for conducting analysis when the data is non normal (other than ADF) in AMOS?

Regards,

Dear Wali,

I would advise to please proceed bootstrapping method. As most of the data gathered naturally is not normally distributed (most of the times) this remedy is considered superior. Byrne (2010) also considered using bootstrapping a superior method.

In AMOS please follow steps as below:

-Go to "Analysis properties"

- Go to bootstrapping

- Check on "perform bootstrap"  put 1000 to number of bootstrapping sample

- Check "Bias-corrected confidence interval" to 95%

- Check on "Bootstrap ML"

- Close the box and run the estimation

In the AMOS output you would be interested in p values reported in "regression weights" and the beta (estimates) reported in "standardized  regression weights". You will need to report these values along with the earlier values that you received without bootstrapping, however you will accept / reject you hypothesis on the ground of these results. 

I hope this answers the question. Best of luck for your endeavors

Thanks. Indeed, it was very useful. I will bother you again for further advice.

Regards,

Wali

You are welcome.

Best of Luck

SAM

Dear S A Moshadi Shah

I'm interested in this rule of thumb that You mentioned:

Check the Critical ration (c.r.) which has to be below the acceptable range of < 5 (Bentler, 2005).

I have not found any Bentler 2005's publication where exactly this value would be given. In other publications (Polish-language) I have seen the threshold = 2. Could you give an exact quote - for example, the name of the article or book?

Thank You and kind regards