I've never heard of this - especially subtracting a set volume (The cylinder size obviously matters). The only correction that I can think of is that the water level in the inverted cylinder is some height above the water level in the pan of water. In that case the pressure in the cylinder is lowered by the effect of weight of the water:
P = Patm - ρgh and the volume has to be corrected using Boyles Law: Vair = Vmeasured (P/Patm)
The correction will be Vmeasured(-ρgh/Patm)
Let's say you measured 20 ml with a height of water of 17 cm. I get the correction to be .3 ml (Interestingly close). Of course it is proportional to the height and the air volume (which oppose each other, so not a lot of variation). I guess this is the reason - I just find the constant -.2 ml weird. (It might be correct for a particular graduated cylinder)
