Everything matters at >GHz.
One good thing to do, is to get some "golden units" of
known pedigree and attributes, that are similar to what
you are working on (LNA for LNA, mixer for mixer, filter
for filter, ...) and determine whether you can get the
known right answer from your bench. Buy top quality
SMA cables that purport to be super low loss well past
where you think you care - if measuring distortion and
intercept points, you have to be good for the harmonics
of interest, not the carrier frequency alone.
Then work your way down the lineup measuring your
and golden units, see where results really start to
diverge and there's where to spend your time on "why?".
That's one of the real tragedies of university education,
hardly ever are you taught what to do when things go
wrong - only theories about how it's be if everything
went right. I think everyone ought to have a "gap year"
where they work as a lab technician in their major focus.
But then, who needs a worthless technician who's going
to leave just before they became useful?