This is the smallest proportion of observations that can make a measure of location (e.g. estimate of central tendency: mean, median, etc) arbitrarily small or large.
In terms of the mean and [OLS-Ordinary Least Squares Regression.doc] this is 1/N; This proportion is the % of data that would need to be altered to throw off estimates. E.g. .3 then 30% of data would need to be altered.