Fitting Johnson distributions using least squares

Abstract
A weighted least squares regression method is proposed for fitting cumulative probability distributions to data. This technique is illustrated for the Johnson translation system of distributions. The least squares procedure minimizes the distance between the vector of uniformized order statistics and its corresponding expected value to identify the Johnson distribution that provides the best fit. This least squares procedure is shown to be numerically robust and to provide a good fit of the data when compared to the empirical distribution. Two examples illustrate the use of the procedure.