Due to its strong requirements in motor abilities, robot table tennis is an important test bed for new robot learning approaches. Learning approaches have to generalize a complex hitting behavior from relatively few demonstrated trajectories, which neither cover all ball trajectories nor all desired hitting directions. Therefore, past approaches that only modeled a deterministic mean behavior without capturing the variability of the movement have been fairly limited. Recent work on capturing trajectory distributions using probabilistic movement representations opens important new possibilities for robot table tennis. In this paper, we present two new methods to adapt probabilistic movement primitives. First we present a method to adapt a probability distribution of hitting movements learned in joint space to have a desired end effector position, velocity and orientation. Subsequently, we present a method to find the initial time and duration of the movement primitive in order to intercept a moving object like the table tennis ball. The resulting methods rely on simple operations from probability theory. Providing a more principled approach to solve some of the challenges of robot table tennis compared to previous approaches. Additionally, the presented method has the potential of generalizing to many other motor tasks.