Weight-frequency data are routinely analyzed either by moment computations or by graphical (percentile) techniques. However, both methods have serious drawbacks. The moment measures are biased, and may have errors induced by grouping or by truncation of part of the sample (pan fraction). The graphical measures may have errors because of interpolation of percentiles. More serious is that the graphical statistics are highly interrelated, so varying amounts of skewness and kurtosis will affect the estimates of all parameters. Because both the moment and graphical methods have defects, an alternative estimation technique is desirable. Fitting a cumulative Edgeworth distribution to the observed cumulative data provides estimates which do not have the above undesirable properties. A brief replication study shows that the variance of the Edgeworth estimates is generally less than that of the graphical and sample moment estimates.

You do not currently have access to this article.