NavList:
A Community Devoted to the Preservation and Practice of Celestial Navigation and Other Methods of Traditional Wayfinding
Re: Rejecting outliers: was: Kurtosis.
From: Marcel Tschudin
Date: 2011 Jan 2, 12:28 +0200
From: Marcel Tschudin
Date: 2011 Jan 2, 12:28 +0200
George, I'm not sure whether I understand your concerns: > But starting with a normal distribution, the resulting scatter in the > median is significantly greater than the scatter in the mean, simply > because so much of the numerical information has been discarded. So Marcel > is paying a heavy price. I wonder if he has made an analysis of any > benefits of this procedure. For any symmetrical distribution mean, median and mode are identical. If you have good reason to believe that the measured data are expected to be symmetrically distributed, the median can be used. A considerable difference between mean and median indicates either the existance of outliers or a the possible existance of a skew distribution. In "normal" datasets there is no a great difference between the mean and the median. In small datasets an outlier contributes too much. One outlier within e.g. 6 data contributes 17% (if not weighted) whereas the outlier may in reality have a much lower probability. The median thus helps to "correct" the influence of outliers in small datasets. Marcel