Changing path coefficients in Fimix PLS

[khauschildt]

Hello everyone,

as a new user of smartpls, I am confused by following phenomenon occuring in my analyses:

When running the FIMIX algorithm several times for the same number of classes on the same model and data, the path coefficients change (sometimes drastically). More specifically, I get the same ones only with class membership reversed once, then new ones (twice).
Why does this happen? How do I decide which ones to use?

I fear that I am missing something really obvious and am sorry for being ignorant. Nevertheless, I would be grateful for even the slightest hint or pointer towards relevant literature.

Thanks in advance!