Significance in MGA

[guru_rr]

Dear Sir
I am running MGA using SmartPLS, and I am finding some strange issue which I will like to understand.
(I am quite new to this PLS SEM thing, so my apology in case my doubts appear naive!!)

I have two groups Experts and Non experts and some relationship constructs in my study where difference between the groups is being explored.
Now in the MGAs I ran, I find minute differences in path coefficients (like 0.02. 0.04) etc getting significant p-values but major coefficient differences ( in order of 0.2,0.6 etc)
have non-significant p-values associated with them!!

Also the individual paths in their own groups are significant and non-significant in two groups as per my hypothesis ( when I run PLS for each group separately I mean) , but strangely the PLS-MGA
as per SmartPLS still says these paths are insignificant!!

Now my question is, if it's possible?

Thanking you for your replies.

Regards
Ramesh

[guru_rr]

Bumping up this question. Would love to hear an answer, too.
Thank you

[jmbecker]

Yes, all of what you describe is possible.

First, the difference between significant and insignificant must not be significant. It is a simple rule that people always confuse.

Second, whether something is significant depends on the size of the coefficient and the associated error. If you coefficient is large, but the error is even larger then it is nonsignificant. If your coefficient is small, but the error even smaller then it is significant. Both are important and together they determine the significance.

[guru_rr]

Thanks Dr. Becker

So just to be clear, when you say

First, the difference between significant and insignificant must not be significant. It is a simple rule that people always confuse.

you mean, the difference between two group's path coefficient can be of any order and be significant or non-significant!
(As in my case 0.02 being significant and 0.23 being non significant)

Also just to be sure, the ERROR in case of bootstrapping would be STDDEV or we should use STDDEV/SQRT(N)?
(Since my two groups have very low N in order of 30 and 40)