AVE or Square Root
of AVE & Significance of Construct Correlations
Hi
Wynne,
1)
AVE or square root of AVE: in the measurement model, should the AVE or the SQRT
of the AVE be larger than the correlations between the construct and all other
constructs in the table. I've seen
papers that use AVE (e.g., Compeau) and I've seen others that use the SQRT of
the AVE (e.g., Igbaria). Which
should it be?
In my book
chapter (edited by Marcoulides), I noted that it should be the square root of
AVE. But if the AVE is already
larger than the correlations, then you can bypass taking the square roots since
those results will automatically be larger than the AVE.
2) significances
of correlations: how can I get the statistical significance of the correlations
between constructs in PLS?
Good
question. At this point in time,
there is no automated solution for calculating the significance
non-parametrically.
The
hard way is to do a bootstrap and physically go thru the boot.lst output file
and write down all the correlations, etc.
I going to
put this down as a to do for the software.
If you use random
sampling and bi-variate normality assumptions, you can calculate the
significance for each correlation by using the following formula:
[r-squared*(N-2)]
/ [1 - r-squared] and looking up
the F table for 1,N-2 degrees of
freedom.