The answer to this question is that it depends on what I am animating.
If I am animating for film, I start in stepped mode and stay in stepped until I get to the polishing phase. This helps me to stay focused on the drawings themselves and not let the in-betweens distract me from focusing on my keys and breakdowns. Once I am done with stepped, I change my curves to linear, then to spline. I pretty much touch every frame and really limit the amount of work I let the computer do because the result it gives me is rarely what I want. I don't touch the graph editor until pretty late in the game. This may mean extra work for me, but I would rather have that level of control if things change late in the process. It's easier for me to go back and make changes without worrying about how the software is interpolating the in-betweens.
If I am animating for games, time is of the essence due to the sheer amount of animation required and for the rapid prototyping of game play. Game engines will also play subframes, so you have to make sure that the rotations won't get into a state of gimbal between frames. In this case you want the computer (software) to do as much as possible. So in this case I start with spline right out of the gate.
Guest Blogger Joe Mandia