In force based models of pedestrian traffic, the relaxation time, ?, is related to the time it takes a pedestrian to adapt its motion to its preferences. An example of this is linear acceleration, but ? is also connected to how the agent adjusts to spatial variations in its preferred velocity, and affects evasive maneuvers. These many roles of ? may be a problem when calibrating force based models. We compare linear acceleration, to new data on, and simulations of, turning movements. The results indicate that the models predict drifting of a magnitude that is not supported by the data.