Page 1 of 1
Significance level for correlations coefficients
Posted: Fri Dec 17, 2010 12:23 am
by Ann Sardesai
Hi,
Can anyone help with how i could determine the significance level or the pvalue for the correlations coeffcients in PLS?
Posted: Sat Jan 08, 2011 1:25 pm
by Diogenes
Hi,
To the
path coefficients, just run the Bootstrap algorithm to have the t-values (t > 1,96 are significant at 5%).
The actual version (2.0.M3) does not compute t-values to the
correlations between LV.
You must copy/paste the LV standardized scores to another software to have these correlations and their probability of significances computed (SPSS in a parametric approach and Resampling for a bootstrap approach -
http://www.uvm.edu/~dhowell/StatPages/R ... pling.html).
Best regards,
Bido
Posted: Thu Mar 03, 2011 9:19 am
by Bjerg54
Hi Professor,
I have a question regarding the t-testing between the LV. If SmartPLS dosnt do this, then what are the results shown in "Path Coefficients (Mean, STDEV, T-Values)" when doing a bootstrap?
You get these results "T Statistics (|O/STERR|)" which I thought were the t-test results of the correlation between the LV's.
Another question regarding the test of a model. When evaluating the model, first I look at the outer model, then the inner model. Is there a meassure of significanse of the total model (outer+inner), not thinking about the r^2.
Kind regards,
Posted: Thu Mar 03, 2011 4:54 pm
by Diogenes
Hi,
Path coefficients are different from correlations.
These results (Mean...) are from path coefficients.
A general measure is the GoF, see:
viewtopic.php?t=1407&highlight=gof
Best regards,
Bido
Posted: Fri Mar 04, 2011 9:53 am
by Bjerg54
Hi Diogenes,
Thank you for the reply.
Yes the path coefficient are as I understand the standardized versions of the linear regression weight results.
When describing the relationship between the LV, and calculating the impact on one LV when increasing another, which measurement would be recommended to use, the correlation or the path coefficient?
BR,
Wilke
Posted: Tue Mar 08, 2011 6:06 am
by mronkko
Actually, you cannot estimate the significance of correlations with SPSS no more than you can estimate the significance of the path coefficients by comparing these to the appropriate t-distribution. The correlation between PLS scores does not follow any known sampling distribution, so you would need to bootstrap the confidence interval.
I do not think that SmartPLS can be used to estimate the significance of the correlations, since you would need to export the construct scores for each bootstrap sample.