The problem of outliers in data has been a topic of lively discussion since the 18th century, when it arose in astronomy, geodesy, calorimetry, and doubtless many other measurements. Let us interpret ‘apparatus’ broadly as meaning any method for acquiring data. On the philosophical side, two opposite views have been expressed repeatedly. (I) Something must have gone wrong with the apparatus; the outlier is not part of the good data and we must throw it out to avoid getting erroneous conclusions. (II) No! It is dishonest to throw away any part of your data merely because it was unexpected. That outlier may well be the most significant datum you have, and it must be taken into account in your data analysis, otherwise you are ‘fudging’ the data arbitrarily and you can make no pretense of scientific objectivity. -- ET Jaynes - Probability Theory, pgs.615--616