The point is, sometimes you do want to test the full environment including the touch screen. For example, you may want to check if the touchscreen locks up after a number of taps, if your OS/input parsing layer can cope with the jitter (a finger, in contrast to MITM-injected events, will almost never hit the same x/y coordinates in a row), or you want to check against real production hardware/software and not against something that's been modified for testing in any way.