故事 · Tukey 与五数概括
Origin Story · Tukey and the Five-Number Summary
1970 年代，普林斯顿的统计学家 John Tukey（也是“比特 bit”“软件 software”这两个词的发明者）想要一种工具：
不画一堆数字、不算复杂公式，就能一眼看穿一组数据长什么样。他给出的答案是五数概括 ——
最小值、Q1、中位数、Q3、最大值，五个数就把一组数据的中心、散布、偏态讲完。
再把它画成一只“箱子加两根胡须”，超出 1.5×IQR 围栏的点单独圈成异常值。
它比直方图更省地方、更适合多组并排，从此成了工程师拿到数据的第一张图。
In the 1970s, Princeton statistician John Tukey — the same man who coined the words bit and software — wanted a tool that could
show the shape of a dataset at a glance, with no tables of numbers and no heavy formulas. His answer was the five-number summary:
minimum, Q1, median, Q3, maximum — five numbers that pin down the center, spread, and skew of a sample.
He drew it as a box with two whiskers, and any point past the 1.5 × IQR fence got flagged as an outlier.
More compact than a histogram and built for side-by-side comparison, the box plot became the first chart engineers reach for.
1 原始队列 → 浓缩成一个箱子 Raw line-up → distilled into one box
无离群No outliers2 箱线图 = 五个数讲完一组数据 The box plot tells a whole dataset in five numbers
下胡须 / Q1 / 中位 / Q3 / 上胡须 = 五数概括。一张图同时给出中心、离散、偏态、异常。
Lower whisker / Q1 / median / Q3 / upper whisker = the five-number summary. One chart shows center, spread, skew, and outliers all at once.
1.5×IQR 围栏:Tukey 定的“正常范围”。超出就是离群点，值得单独追查根因。
1.5 × IQR fences: Tukey's definition of "normal range". Anything beyond is an outlier worth a root-cause hunt.
中位偏向一侧：箱子里中位线靠左=右偏，靠右=左偏 —— 不用算偏度也能看出来。
Off-center median: median line pushed toward the left edge = right-skewed; pushed right = left-skewed. No skewness formula needed.
抗异常：箱子基于分位数，拖再远的离群点，Q1/Q3 也几乎不动（对比红三角均值被拉跑）。
Robust: the box is built on quantiles, so Q1/Q3 barely budge no matter how far you drag an outlier — unlike the mean (red triangle), which gets yanked along.
3 现实里的箱线图 Box Plots in the Real World
五数概括：拿到一组数据先画箱线图，5 秒读出中心（中位）、散布（IQR）、偏态、异常，无需任何公式。
Five-number summary: When data first lands on your desk, draw a box plot. In 5 seconds you have center (median), spread (IQR), skew, and outliers — no formulas required.
异常值检测：1.5×IQR 围栏自动圈出可疑点 —— 量错、设备异常还是真离群？那往往是改进的突破口。
Outlier detection: the 1.5 × IQR fence circles suspects automatically — bad measurement, equipment glitch, or a real anomaly? That's often where the improvement lives.
多组对比：各班身高、各机台尺寸、各供应商批次，多个箱子并排，散布与中心差异一眼可比。
Side-by-side comparison: heights by class, dimensions by machine, batches by supplier — drop the boxes next to each other and the spread-and-center gaps jump out.
Tukey 的遗产：探索性数据分析（EDA）的标志性工具，统计软件、SPC、过程能力研究里随处可见。
Tukey's legacy: the signature tool of exploratory data analysis (EDA), and a fixture in every stats package, SPC dashboard, and capability study.
一句话In One Line
箱线图是工程师的“第一眼工具”：拿到一组数据先画箱线图，5 秒钟就能看出中心、散不散、偏不偏、有没有异常。
多组并排还能一眼比较（各班身高、各机台尺寸）。它最大的价值是自动把“值得追查的异常点”圈出来 —— 那往往就是改进的突破口。
The box plot is the engineer's first-glance tool: draw one and in five seconds you know
where the center is, how wide the spread is, whether it's skewed, and whether anything weird is going on.
Line several up side by side and comparisons are instant. Its biggest payoff is that it
automatically circles the points worth investigating — which is usually where the next improvement is hiding.
常见误用Common Mistakes
看到离群点就直接删。离群点是线索不是垃圾，先查根因（量错？真异常？）。
Deleting outliers on sight. Outliers are clues, not noise. Investigate root cause first (mis-measured? real anomaly?).
样本太少（n<5）还画箱线图。点太少分位数不稳，直接看散点。
Drawing a box plot when n < 5. Quantiles are unstable with so few points — just plot the raw scatter.
只看箱子忘了原始点。叠加抖动散点，避免双峰/断层被箱子掩盖。
Reading the box without the raw points. Overlay a jittered scatter — bimodal humps or gaps hide easily inside a single box.