It is sometimes necessary to quote the error variance, or *mean square error* of scalar transformed responses, such as reaction times, that have very large or small values. In this case the mean square error for the *raw* data may be obtained from the mean square error for the transformed data by simply multiplying the transformed data's mean square error by the square of the inverse transform.

For example let us suppose the data is divided by 1000 and a mean square error for this transformed data is found to be equal to 0.0125. The mean square error of the original data is, therefore, equal to 1000^{2 } times the mean square error of the transformed data = 10^{6 } x 0.0125 = 12500.