A Simple Alternative to Student's T

An easily computed t‐statistic is defined in terms of a trimmed mean and a highly efficient linear estimator of scale. It is shown that the statistic is satisfactorily approximated by a multiple of a Student's t variable with slightly reduced degrees of freedom. Alternatively the statistic may be referred to a table of percentage points produced by an efficient Monte Carlo technique. The procedure requires only a partial ordering of the data into three groups and the calculation of the averages of the observations in these three groups. No sums of squares are involved. For observations from a normal distribution the procedure is shown to have power only slightly lower than that of t whereas for certain long‐tailed non‐normal situations it is demonstrated that the alternative has greater power than t to detect moderate to large shifts in population mean.