Tests for Departure from Normal in Laboratory Data

This study was undertaken to compare several methods of evaluating the common assumption that laboratory data may be treated as if they had been drawn from an underlying Gaussian distribution. As had been described earlier, the usual Kolmogorov-Smirnov test was found to be unsatisfactory due to its sensitivity to rounding. In addition, a recently suggested modification also was shown to be unsuitable, because it was quite insensitive when applied to simulated data known to be positively skewed. The authors suggest the following: (1) that the coefficients of kurtosis and skewness are valuable for use against specified alternatives; (2) that the traditional Kolmogorov-Smirnov test is appropriate if the degree of rounding is minimal; and (3) that a satisfactory “omnibus” test is not yet available.