故事 · 比尔·盖茨走进酒吧
Story · Bill Gates Walks Into a Bar
统计学界有个流传已久的段子:比尔·盖茨走进一家小酒吧,他一进门,屋里所有人的「平均资产」瞬间暴涨到上亿 ——
可大家口袋里的钱一分没多。同一刻,「中位数」纹丝不动,依旧是那个普普通通的数字。
这一幕揭示了一个反复被忽略的真相:均值会被极端值狠狠拉爆,中位数却稳如泰山。
均值、中位数、众数,这三种「中心」脾气完全不同。它们都自称代表「典型」,可一旦人群里混进一个盖茨,谁还值得信,立刻见分晓。
A classic statistics joke: Bill Gates walks into a small bar and the moment the door swings, the "average net worth" of everyone inside shoots into the hundreds of millions —
yet not a single wallet got heavier. In the same instant, the median does not budge — still the same ordinary number.
That snapshot lays bare a truth people keep forgetting: the mean gets dragged by extremes; the median sits like a rock.
Mean, median and mode are three very different "centres". All three claim to summarise the "typical" person — but the moment a Gates wanders in, you instantly see which one you can still trust.
平均年收入 MeanMean income
5.0 万
所有人的钱倒进一个桶,搅匀平分pool everyone's money and split evenly
中间那个人 MedianMedian (the middle person)
5.0 万
按收入排队,站正中间那位 = 普通人真相line everyone up by income — the one in the middle is the honest "typical"
1 这间酒吧里坐着谁 Who's Sitting in This Bar
岁月静好all quiet10 个普通人,收入差不多。屏幕顶上「平均」和「中间那个人」几乎一样 —— 现在按左边的红按钮,看会发生什么。 Ten ordinary regulars with similar incomes. "Mean" and "middle person" up top read almost the same — now hit the red button on the left and watch what happens.
2 真相:把所有人按收入排成一条线 The Truth: Line Everyone Up by Income
均值≈中位mean ≈ median绿线 = 中位数:站在普通人正中间,雷打不动。 红三角 = 均值:盖茨一进来,它就被拽离人群,飞进一片一个人都没有的荒野 —— 它谁也不代表。 Green line = median: planted right in the middle of the regulars, unmovable. Red triangle = mean: the moment Gates arrives it gets yanked off the crowd and into empty wilderness where nobody lives — representing no one.
现实里,到底信哪个数In the Real World, Which Number Do You Trust
收入统计:「人均可支配收入」用均值,被高收入拉高;想看普通人真实水平,必须查「中位数收入」。
Income statistics: "per-capita disposable income" uses the mean and is pulled up by high earners. If you want the honest level of the typical household, look at median income.
房价中位数:楼市报告几乎都报中位数房价,正是因为均值会被几套豪宅天价拽飞,中位数才反映「典型一套房」。
Median house price: housing reports almost always quote median price — a few mansions yank the mean far above reality, while the median reflects the "typical home" actually changing hands.
交付/响应时间:工期、客服响应这类右偏数据,几个超长拖延会把均值抬高;承诺客户该看中位数或 P90,而非平均。
Delivery and response times: lead times and customer-service responses are right-skewed — a few long drags inflate the mean. Quote customers the median or P90, never just the average.
异常值影响:一个录入错误(多打一个 0)就能毁掉整组均值;中位数对异常值几乎免疫,是稳健统计的首选。
Outlier resistance: a single typo (one extra zero) can wreck an entire mean. The median is nearly immune to outliers — the first choice in robust statistics.
一句话 · 你该信哪个数In One Line · Which Number Should You Trust
老板汇报「平均工资」The boss quotes "average salary"
爱用均值 —— 被少数高管/老板拉高,显得人人收入都不错。
Loves the mean — pulled up by a few executives or owners so the room looks well paid.
「我这种普通人拿多少」"What does someone like me actually earn?"
看中位数 —— 不被富豪拽跑,才是真实的「典型人」。
Look at the median — the tycoons cannot drag it, so it actually represents the typical person.
口诀:对称用均值,偏态用中位,类别用众数。 只要数据里可能藏着「盖茨」(长尾极值),就别信平均。
Rule of thumb: symmetric → mean, skewed → median, categorical → mode. The moment a "Gates" (long-tail extreme) might be hiding in the data, do not trust the average.
常见误用Common Mistakes
收入/房价/工期这类右偏数据报均值 → 被极值拉高失真。用中位数,或先做对数变换。
Reporting the mean for right-skewed data like income, house prices or lead times → extremes drag it off-truth. Use the median, or log-transform the data first.
给类别数据算均值(「满意度颜色平均=3.7」没意义)。类别只能用众数。
Averaging categorical data ("average satisfaction colour = 3.7" is meaningless). Categories can only have a mode.
只报中心、不报离散。集中趋势必须和标准差/IQR 一起看(下一页)。
Reporting a centre with no spread. Central tendency always travels with standard deviation or IQR (see the next page).