0.4 to 0.8%, two to four times PLA's contraction. A 100 mm dimension finishes somewhere between 99.6 and 99.2 mm.
The number itself is manageable. What makes ABS difficult is that the same contraction has to happen while the part is still stuck to a hot plate.
Shrinkage and warping are one phenomenon
An unconstrained ABS part would simply come out slightly small. A printed one cannot contract freely, because its base is welded to the bed and its lower layers have already set.
So the contraction of each new layer pulls on everything below it, and the accumulated stress goes to whichever bond is weakest — the plate, giving a lifted corner, or a layer interface, giving a crack partway up a tall part.
This is why ABS requires an enclosure while PETG only benefits from one. The enclosure is not preventing shrinkage; it is slowing the gradient so the stress never peaks.
Compensating for it
Across 100 mm, model the dimension at 100.4016 mm at the bottom of the range or 100.8065 mm at the top — corrections of 0.4016 and 0.8065 mm respectively.
Which end of the range you are on depends on your chamber temperature, so it genuinely varies between machines and between seasons. Measuring one printed test part is worth more than picking a figure.
Why one scale factor will not do
Contraction is not isotropic and, on ABS, the anisotropy is large enough to see. X and Y contract far more than Z, which is restrained by the layer stack.
A part scaled uniformly by 100.6% will come out roughly correct across the plate and noticeably tall. Compensate X and Y; leave Z.
The filled version is different
ABS-CF contracts only 0.2 to 0.5% — no worse than PLA — because the chopped carbon fibre restrains the polymer matrix as it cools. That dimensional stability, not stiffness, is the main reason people run filled ABS for large flat parts.
It still needs the enclosure. It just fights it less.
Compute a correction in the shrinkage calculator, and see the material's full profile on the ABS page.