The empirical evidence presented in this paper indicate that for the four cases examined, the standard error of measurement was not constant throughout the score range, as is usually assumed, but rather varied considerably, being relatively higher for scores in the middle of the range and lower for scores in the upper and lower portions of the range. The previously derived equation when applied to the multiple-choice data was found to reflect the curvilinearity in the error of measurement data, but to give a consistent underestimate of the error for scores close to the chance score range. A better representation of the error data seemed to be given for the more homogeneous figure classification test data than for the O.S.P.E. data. Some adjustment of the formula or of its application would seem to be advisable to allow it to give a better representation of the error of measurement at various parts of the score range on a multiple-choice test.
