Definition and Detection Methods
An outlier is an observation that lies significantly far from the bulk of a dataset. Detection methods include flagging values more than 1.5 times the interquartile range (IQR) below the first quartile or above the third quartile, or those more than three standard deviations from the mean.
Impact on the Mean
Outliers can distort the mean. If one person with a $1 billion income joins a group of ten, the average income becomes wildly unrepresentative. This is why income statistics also report the median as a measure of central tendency.
However, carelessly removing outliers risks discarding important information. Distinguishing measurement errors from genuinely extreme values requires careful judgment.
Robustness of Percentiles
Percentiles are rank-based statistics and therefore nearly immune to outlier effects. No matter how much the top 0.1% increases their wealth, the 50th percentile (median) remains unchanged. This robustness is one advantage of using percentiles for rankings.
Outliers and Ranking Metrics
Because the ranks of other values remain unchanged however extreme the maximum value becomes, rank-order statistics such as percentiles are used to compare heavy-tailed data such as income. Noticing whether a figure is mean-based or rank-based is one clue for reading it without being swayed by extreme values.