Learning curves: Facts or artifacts?

"It is our purpose in this paper to show that the arbitrary choice of a cutoff point in the dichotomizing of continuous response distributions can impose significant constraints upon the shape of resulting learning curves, and that this can form the basis of misleading theoretical interpretations. We have chosen for illustration of this point the use of time-on-target scores as indicants of the level of skill attained in tracking tasks. However, we believe that the principles developed are quite general and apply to many learning situations." 18 references.