Gps antenna location

I have a Gps rx, neoblox6m I use it as part of a frequency reference.
Since I now have a means of charting the accuracy of the reference I've noticed that the accuracy depends a lot on the location of the antenna.
Directly under a glass skylight its horrible, around 100 part per billion, under a concrete tiled roof only 6 feet awat its around 20.
I didnt expect it to be so different, do I really need to make an outdoor enclosure and stuff this thing outside?
Oh the measurements were all made with the same setup, nothing moved or changed, so I'm assuming the accuracy change is the antenna.

I'd suspect some sort of multi-path effect; something reflecting a fairly strong signal with the antenna in the bad location, so messing up the apparent path length?

I have the GPS antenna for the one I use as an NTP ref clock outside on the roof - it's small enough to lift a tile slightly from inside the loft and slide it through.
(Mine's based on a GPS25 module, set up many years ago..)

Yes I spose multipath is an issue with short wavelengths.