Its not obvious so I'll try to explain.
Lets say we have two transformers (any for the moment save for they are the same).
Also the transformer have a separated primary and secondary as that makes it easier to see.
So we hook them up back to back and hook the primaries to the tracking generator and the
spectrum analyser input. So far we have complete circuits that do not depend on any
circuit common or "ground".
For that case if I said the transformers did such and so we can see and evaluate it.
Now what if... I break the back to back connection for one lead and insert 2M
of wire nicely laid out on the bench would you be happy with that? I know I would
not as that 2M of wire is about 2.6uH of inductance give or take. So I'd expect the
trace to show some loss especially at 28mhz especially since that 2.6Uh is about
480 ohms of reactance. Please do not argue the amount of inductance its
magnitude is less important than the fact that its there at all.
So if I go back to the back to back original transformer and instead take the other
leads apart and connect them to the input side shield of the coax what have we?
THe secondary are connect together via a 2M lead that also includes the frame
of the SA. But the SA is ground.... no it has a common, but ground is likely about
3.5ft under it or more and the lead getting it to that mythical point is 3.5ft long or more
with its inductance. IF you go to the 3 wire cord its 3-6ft long to the safety ground
pin on the receptacle which is not RF ground so it does not help and likely adds
problems to ponder. So as we follow the current around the circuit we find we have
a 2M loop again and that loop has an added unknowns in the middle.
Its very easy to do this and not realize there is more going on but it is far from complicated
only that there are many parts to it. Solution is to fix the test to remove that potential error.