Critique of methods of estimating the variability of metabolizable energy from assays with fasted roosters

Metabolizable energy values from assays with fasted roosters are often calculated: 1) without regard to variability in data from control roosters, 2) with chemical determination on pooled and not individual data, and 3) from data sets with statistically aberrant points removed. Data from determinations of the ME of a feather meal sample were used to examine the influence of these practices on the reported mean and variance values. Ignoring the variability of the fasted controls or pooling excreta samples did not have large effects on the means or variances of these samples. Eliminating data from roosters whose excreta weights are more than one standard deviation from the mean had little effect on the mean value but made the variance appear much smaller than it actually was. As a result of these practices, researchers may place too much confidence in their data and fail to use enough replicates (roosters) to justify that confidence.