Argument for IQ Analysis

The argument is pretty simple - means work well for describing trends of normal distributions, and my distributions are not normal. Take, for example, the distribution of NDCGs for L4 at the 0.9 convergence threshold...
Clearly, the distribution is not normal, and has a very long tail. When measuring skew across all thresholds (below), we see that this tendency holds as the majority of the distribution meets the performance ceiling. Hence, an interquartile mean analysis is more appropriate.