Median

The median is the middle value in a dataset ordered from smallest to largest — for an odd number of values, the single value exactly in the center; for an even number, the average of the two middle values. Unlike the mean, the median's calculation depends only on the position of values within the ordered sequence, not on their actual magnitude, which is the specific property that makes it resistant to outliers in a way the mean fundamentally is not: a value doesn't need to change the median at all just by becoming more extreme, as long as it stays on the same side of the middle.

This resistance to outliers, formally called robustness, is why the median is the preferred measure of central tendency for skewed distributions — income, home prices, wait times, and other quantities where a small number of unusually large values can exist alongside a much larger number of typical ones. Household income is the standard illustrative case: a dataset where most households earn around $60,000 and one earns $5 million will show a mean pulled dramatically upward by that single value, while the median stays close to $60,000 regardless of how extreme that one outlying household's income actually is — which is exactly why government statistical agencies like the U.S. Census Bureau report median household income rather than mean household income as the primary figure, specifically because it more accurately reflects what a typical household actually earns.

The median is also the one measure of central tendency rigorously valid on ordinal data, where values have a meaningful order but no fixed, equal distance between them — a 5-point satisfaction scale's median response is a defensible, technically sound summary, while computing a mean on the same scale implicitly (and not entirely rigorously) assumes the gaps between "poor," "fair," "good," and "excellent" are all equal, an assumption the median never requires. This is a direct consequence of the median depending only on order, not on magnitude or spacing — the same property that gives it outlier resistance on numerical data also makes it the statistically correct, rather than merely convenient, choice of central-tendency measure on ordinal data.

The mean-versus-median comparison itself carries diagnostic information about a distribution's shape: when a distribution is symmetric, the mean and median are approximately equal, but when a distribution is right-skewed (a long tail of unusually large values, as with income), the mean sits noticeably above the median, pulled upward by those large values, and when a distribution is left-skewed, the relationship reverses — which is why reporting both the mean and median of a dataset together, rather than either one alone, is a simple but genuinely informative diagnostic for spotting skew without needing a full histogram.

Computing a median requires sorting the entire dataset first, which is computationally more expensive than computing a mean (a simple running sum), a difference that mattered more in earlier decades of limited computing power and is largely negligible at modern data scales for all but the very largest datasets, though it remains the technical reason some streaming or real-time systems default to tracking a running mean rather than a running median, since an exact median generally can't be updated incrementally as cheaply as a mean can.

Read the median as the central-tendency measure defined by position rather than magnitude, and that single structural difference from the mean is the source of everything distinctive about it: its resistance to outliers is why it's the standard choice for skewed real-world quantities like income, its validity on ordinal data follows directly from depending only on order rather than assumed equal spacing, and the gap between a dataset's mean and median is itself one of the simplest, most informative signals of how skewed that dataset actually is.