We generally adjust caster by tilting the top of the strut back. The strut is directly bolted to the spindle, so that means as we tilt the strut back we're also tilting the spindle back the same amount. The tie rods attach to the steering arm, which sticks off the front of the spindle. Hopefully it's not too hard to visualize that as you tilt the spindle back, the steering arm will come up. Meanwhile, we haven't adjusted the position of the steering rack at all, so the position of the steering arm has now changed relative to the steering rack, which means all those things slowgo said could be affected.
Bumpsteer is the tendency of the wheels to toe in or out as the suspension moves through its range of motion. As the control arm moves through its arc, if the tie rods aren't the right length they'll move through a slightly different arc, which means the spindle will be turned in our out. Does that make sense? This can be corrected by using spacers so that the control arm and tie rods scribe the same arc. Here's a picture of the Griggs Racing front suspension on my Mustang (also a MacPherson strut setup, like the Subie) - in front of the yellow strut you can see the end of the steering arm poking out from behind the rotor, and then about 2-1/2" below that you can see the tie rods, with a nice beefy -AN spec bolt and a spacer stack to line up the geometry correctly.
http://www.submariner.org/thepno95/P...0K-member5.jpg
Roll steer is similar, except now the suspension is moving through it's motion due to the roll of the car - one side goes into compression, the other side into rebound.
Ackermann, well, I've never really understood that one. I just know it's better on my Mustang with the newer '94/95 spindles and the Griggs stuff.
Pat Olsen
'97 Legacy 2.5GT sedan