Research work · 1970
Interpretation of the Kurtosis Statistic
Brad S. Chissom
The American Statistician
Abstract
A description of the kurtosis statistic has long been overlooked by authors in statistics and measurement. This study illustrates how the kurtosis statistic may be correctly interpreted when it is computed for approximately normal, rectangular, and bimodal distributions of measures. The results of the study point out that in rectangular distributions the kurtosis value is approximately −1.20 to −1.25. Other results point out that the number of cases in the tails of the distribution drastically affect the kurtosis value. Truncation in the tails of a distribution may cause the kurtosis value to change from positive to negative with only the deletion of a small number of cases. Finally, the results point out the fact that a perfectly bimodal distribution of measures has a kurtosis value of −2.00.