Correct order of the connector´s nodes

Hello Everyone,

I would like to kindly ask a question (maybe a trivial one, but...).
I have a simplified model with a four CONN3D2 (Connector Section CARTESIAN, CARDAN) connectors at mounting point, simulating bushings. 

One node of each connector is connected to the RBE2 rigid element (pink on the picture). Master Of this RBE2 is the center of gravity of the simplified system (with concentrated mass) and I have a several Load cases, where I apply acceleration to this mass in different directions - Searching for the reaction forces in the grounded nodes of the Connectors. 

The second node of each connector (node id1 in the case marked with dotted line on the picture) is fully grounded.

The connector in the model has ZERO length - meaning that node id2 has the same coordinates as node 1.

Each connector has a given stiffness curves for all it´s components (1 to 6) and has dedicated local reference (RECTANGULAR) system for these stiffness  (with origin having coordinates same as a grounded node of the connector  - node 1 for example) 

Now the question is about CONN3D2 element:

How does the order of the connector´s nodes affects the results? 

I mean:

*ELEMENT, TYPE=CONN3D2, ELSET=Conn_UP_FL
10000001, 1, 2

or

*ELEMENT, TYPE=CONN3D2, ELSET=Conn_UP_FL
10000001, 2, 1

Depending on the Load case, there differences between two (in reaction forces I measured) are varying from none/negligible to quite significant , if I compare these two ways how to create the connector.

In fact, I read somewhere that the grounded node should be set as first, because the second then behaves relatively to the first.. But I would like to understand more deeply the background, how does it work in this specific case which I described above.

Is there any strict rule in such a case? Because for me,  the both versions do work and sometimes even with almost the same results, and sometimes with quite different. And sometimes the results obtained from the second possibility (so the not-constrained node choosen as first) are more similar to results which I´m comparing them with (from another software)

Thank you very much for any answer.