Re: Data Analysis: An old problem revisited
Howard Shapiro (hms@shapirolab.com)
Wed, 12 Feb 1997 21:39:23 -0500 (EST)
The median is a more robust statistic (i.e., less sensitive to outliers and
asymmetry of distributions) than the mean, so it makes sense to compare
medians rather than means. The problem with taking a ratio is that your
control values are too small; single digits on an RFI scale of 1-10000 are
likely to represent measurements dominated by noise. Differences between
medians on a log scale are, in effect, ratios, and present the same
problem. The difference between medians on the linear RFI scale is probably
a safe measure.
-Howard