The usual explanation is that ABS shrinks more than PLA. It does. It is also nowhere near a sufficient explanation, and believing it is why people attack warping with brims and bed temperature and get nowhere.
The contraction difference, measured
Take a 250 mm edge — a decent-sized enclosure panel — and let the material contract freely at the top of its stored range.
PLA, at 0.5%, ends up at 248.75 mm. It wanted to be 1.2563 mm shorter than you drew it.
ABS, at 0.8%, ends up at 248 mm, wanting 2.0161 mm. At the bottom of ABS's range, 0.4%, it is 249 mm and 1.004 mm.
So across the stored ranges ABS contracts between about 1.6 and 2 times as much as PLA. Real, but bounded: on that panel the difference between the two worst cases is under a millimetre of free contraction.
Under a millimetre. And yet one of those panels comes off the plate flat and the other lifts its corners clear of the plate and cracks along a layer line partway up. The contraction figure is not carrying that.
What the number leaves out
Warping is not caused by contraction. It is caused by contraction that happens while the part is too stiff to relieve it.
Every layer is deposited hot and immediately starts pulling inward as it cools. Underneath it is material that has already cooled and does not want to move. If the new layer is still soft while that mismatch develops, it simply flows and the stress evaporates. If it has already gone rigid, the mismatch is stored as tension, layer after layer, until the accumulated moment at the corners exceeds whatever the first layer's grip on the plate can resist.
So the question is not "how much does it contract" but "how much of the contraction happens after it has stopped being able to relax".
Look at where each material becomes structurally rigid. The heat deflection figures here put PLA at 52–60 °C and ABS at 88–100 °C. Heat deflection is not the glass transition and should not be read as one, but as a stored proxy for the temperature below which this part behaves as a solid it puts the two materials in the right order and roughly the right ratio.
In a room at ordinary temperature, an ABS part is rigid for something like seventy degrees of its cooling. A PLA part is rigid for barely half that. Multiply a contraction coefficient that is roughly double by a rigid interval that is also roughly double, and the compounding is what you actually see on the plate. That is why the difference in outcome is so much larger than the difference in the shrinkage percentages.
Polycarbonate proves it is not just shrinkage
If shrinkage percentage were the whole story, polycarbonate at 0.5–0.8% would warp about like ABS at 0.4–0.8%. It does not. PC produces the most violent corner lift of any common filament, and it does so with a shrinkage range that is no worse.
The extra factor is stiffness. The same strain in a stiffer material generates more force, and PC is much stiffer than ABS. Warping is a force problem, and force is strain times modulus — a material can be dimensionally well-behaved and still tear itself off the bed.
That is also why ABS-CF exists. The chopped fibre halves the contraction and restrains it along the extrusion direction, and warping largely goes away — even though the fibre makes the part stiffer, which by the argument above should make things worse. It does not, because the contraction term fell further than the stiffness term rose. The filled versus unfilled comparison has the rest of the tradeoff.
What actually fixes it, in order of effect
Chamber temperature. If the failure mode is contraction happening after the part goes rigid, the fix is to keep the part above that point for as long as possible. Raising the ambient air around the print does exactly that. This is why an enclosure is a requirement for ABS rather than a nicety, and why the answer to whether you need one is not size-dependent in the way people hope.
Reducing the contraction itself, by using a filled grade. Second-best only because it changes the material.
Geometry. Round the corners, add fillets where the panel meets a wall, and avoid large uninterrupted flat bottoms. Corners lift first because that is where the accumulated in-plane tension has the shortest path to the edge.
Adhesion. Brims, glue, textured plates. Note the position on this list. Adhesion does not reduce the stress; it just raises the force the joint can survive before it lets go — and if the joint wins, the stress has to go somewhere else, which is how a well-glued ABS part cracks along a layer instead of lifting. Both failures have the same cause. The full symptom checklist is on the warping page, and bed adhesion surface by surface covers the plate side.
Bed temperature last, and only within the material's own window. A hotter bed keeps the lowest few layers soft; it does very little for a part that is 200 mm tall.
Why PLA is not actually immune
PLA warps too. It just needs help.
A large flat part, a cold draught from a window, a cooling fan at full tilt on the first few layers, or a bed running at the bottom of its 0–60 °C window will all produce corner lift in PLA. The mechanism is identical; it simply takes more provocation, because both multipliers are smaller.
The practical tell is that PLA warping is nearly always environmental and ABS warping is nearly always structural. If PLA lifts, something changed in the room. If ABS lifts, nothing changed — you were always going to be fighting it.
If you have no enclosure
Two honest answers, and one that gets recommended wrongly.
ASA is not the answer. It is stored at the same 0.4–0.8% as ABS with essentially identical behaviour, and it is chosen for UV resistance, not printability. It will warp on your open frame exactly as ABS does.
The workable substitutions are a filled styrenic if you must have the heat resistance, or stepping sideways to a copolyester if you do not. That decision is a material-requirements question rather than a warping question, and PLA against ABS is the honest starting comparison — most parts specified in ABS are specified there by habit.
And whatever you end up printing, if the part has to fit something, the contraction still has to be compensated even when it does not warp. That is a separate decision made per dimension, and the compensation calculator turns the percentage into the number you model at.