Visual walkthrough — Measures of dispersion — variance, standard deviation

We use one running dataset the whole way: the marks $3,5,5,7,10$ (out of 10). Watch these five dots move through the whole story.

---

Step 1 — Put the data on a line

WHAT. We take five numbers and place each as a dot on a horizontal line (an axis). The axis is just a ruler: position = value.

WHY. Before we can talk about "spread," we need to see the raw data. Spread is a fact about how these dots are arranged — you cannot measure it in your head from a list, but you can see it on a line.

PICTURE. Five dots. Some are close together (the two 5s sit on top of each other), one straggler sits far out at 10.

---

Step 2 — Find the centre (the mean)

WHAT. We compute one special number, the mean, and mark it on the same line.

$\bar{x}=\frac{1}{n}{\sum }_{i=1}^n{x}_i=\frac{3+5+5+7+10}{5}=\frac{30}{5}=6$

Reading it term by term: $\sum _{i=1}^n{x}_i$ is "add up all the dots" $=30$; the $\frac{1}{n}=\frac{1}{5}$ divides by how many there are; $\bar{x}$ (read "x-bar") is the result, $6$.

WHY. Spread means "spread around something." The mean is that something — it is the balance point, the spot where the line would balance if each dot were a coin of equal weight. (See Mean, Median, Mode — measures of central tendency for why it balances.)

PICTURE. A downward triangle (a see-saw pivot) sits at $6$. The far dot at $10$ tugs the balance rightward exactly as much as the two low dots at $3$ and $5$ tug left.

---

Step 3 — Measure each dot's distance from the centre

WHAT. For every dot we draw the arrow from the centre $\bar{x}$ out to that dot. Its signed length is the deviation:

$d_i=x_i-\bar{x}$

Here $\bar{x}=6$, so the deviations are $3-6=-3,5-6=-1,5-6=-1,7-6=+1,10-6=+4.$

Each symbol: $x_i$ is where the dot is, $\bar{x}$ is the centre, and their difference $d_i$ is "how far, and which way." A minus means the dot sits left of centre; a plus means right.

WHY. Distance-from-centre is the raw material of spread. A dot far from centre (the $+4$) contributes a lot; a dot near centre (the $\pm 1$s) contributes little.

PICTURE. Orange arrows point left (negative) from the low dots, teal arrows point right (positive) from the high dots. Length = how far.

---

Step 4 — Why we can't just average the arrows

WHAT. Naive idea: "average the deviations." Let's try: $\sum d_i=(-3)+(-1)+(-1)+(+1)+(+4)=0.$

WHY this fails. The left-pointing arrows and the right-pointing arrows have equal total length — that is exactly what "balance point" means from Step 2. So they cancel to zero every single time, no matter how scattered the data is. A spread measure that is always $0$ tells us nothing.

PICTURE. The orange (left) arrows stacked end-to-end reach the same total length as the teal (right) arrows — a visual tug-of-war ending in a perfect tie at $0$.

---

Step 5 — Square each deviation to kill the signs

WHAT. Replace each arrow by a square built on it. A square needs a positive side length, so we take the side to be the absolute deviation $|d_i|$ (the arrow's length, sign discarded). Its area is then $\text{side}=|d_i|\Rightarrow \text{area}=|d_i{|}^2=d_i^2,$ and numerically $(-3{)}^2=9,(-1{)}^2=1,(-1{)}^2=1,(+1{)}^2=1,(+4{)}^2=16.$

Why is $|d_i{|}^2$ the same as $d_i^2$? Because $|d_i{|}^2=(\pm d_i{)}^2=d_i^2$ — squaring a length already erases the sign, so the side length $|d_i|$ and the raw $d_i$ give the same area. The little ${}^2$ means "multiply the number by itself," and $(-3{)}^2$ and $(+3{)}^2$ are both $9$: squaring throws away the sign but keeps the size.

WHY squaring and not absolute value? Two ways kill signs: $|d_i|$ and $d_i^2$. We pick squaring because (a) it is smooth — a square's area changes gently, so we can do calculus on it later (this is why Least Squares Regression uses squares); (b) it punishes big misses harder — the $+4$ becomes a whopping $16$, dwarfing the $\pm 1$s that become just $1$; (c) squared distance is the geometry that the Normal Distribution is built from.

PICTURE. Each arrow of length $|d_i|$ grows into a literal square of that side. The $+4$ deviation's square ($16$) visibly dominates the tiny unit squares.

---

Step 6 — Average the squares → variance

WHAT. Add all the square-areas and divide by how many there are: ${\sigma }^2=\frac{1}{n}{\sum }_{i=1}^n(x_i-\bar{x}{)}^2=\frac{9+1+1+1+16}{5}=\frac{28}{5}=\mathrm{5.6.}$

Term by term: $(x_i-\bar{x}{)}^2$ is one square's area; $\sum$ adds all five areas $=28$; $\frac{1}{n}=\frac{1}{5}$ turns the total area into the average area; ${\sigma }^2$ (read "sigma-squared") is that average, $5.6$.

WHY average, not just sum? A plain sum grows if you add more data points, even data that's equally spread — that's not fair. Dividing by $n$ gives the typical square-area per point, a number that describes the spread itself, not the sample size.

PICTURE. All five squares are melted together and re-poured into one big square whose area is the average, $5.6$ — the "typical square" that stands in for the whole dataset.

---

Step 7 — Square-root back to real units (standard deviation)

WHAT. Variance is $5.6$ marks² — a squared unit, awkward to talk about. Take the square root: $\sigma =\sqrt{{\sigma }^2}=\sqrt{5.6}\approx 2.37\text{marks}.$

The ​ undoes the squaring of Step 5: it turns an area back into a length.

WHY. We want to answer "how far, typically, is a mark from the average, in marks?" Only a length can answer that. $\sigma \approx 2.37$ means: on average the marks sit about $2.37$ away from $6$. That is a sentence a human can use.

PICTURE. The big "average square" of area $5.6$ has its side length drawn back onto the original number line: a band of width $\sigma \approx 2.37$ on each side of the centre $6$, catching the typical dots.

---

Step 8 — The degenerate case: no spread at all

WHAT. Suppose every mark is the same: $6,6,6,6,6$. Then $\bar{x}=6$, every deviation $d_i=6-6=0$, every square is $0$, so ${\sigma }^2=\frac{0+0+0+0+0}{5}=0,\sigma =0.$

WHY it matters. This is the only way variance can hit zero, and it's the sanity floor: variance is never negative (it's an average of squares, and squares are $\ge 0$), and it equals $0$ exactly when all values are identical. No spread ⇒ no arrows ⇒ no squares ⇒ zero.

PICTURE. All five dots pile onto the single point $6$. Every arrow has length $0$; there are no squares to draw.

Recall Check the edge case yourself

Data $4,4,4$: mean is 4, all deviations are 0, so the variance equals $0$, and the SD $=0$ too.

---

The one-picture summary

Every step in a single frame: dots → centre → arrows → squares → average square → square-root band. Follow the numbers $3,5,5,7,10\to \bar{x}=6\to d=-3,-1,-1,1,4\to d^2=9,1,1,1,16\to {\sigma }^2=5.6\to \sigma \approx 2.37$.

Recall Feynman: tell the whole walk in plain words

Line up your five test scores on a ruler. Find the balance point — that's the average, 6. From the balance point, draw an arrow to each score: some point left, some right. If you just added those arrows they'd cancel to nothing (that's what "balance point" means), so instead you turn each arrow into a square tile whose side is the arrow's length — now left and right can't cancel, and the far-out score makes a huge tile. Average all the tiles into one "typical tile"; that average area is the variance (5.6). Its side length — take the square root — is the standard deviation (about 2.37), which finally answers in plain marks: "a typical score sits about 2.37 away from 6." And if all your scores were identical, there'd be no arrows, no tiles, and the spread would be exactly zero.

---

Connections

• Measures of Dispersion — variance, standard deviation — the parent note this walkthrough builds.
• Mean, Median, Mode — measures of central tendency — why the mean is the balance point of Step 2.
• Normal Distribution — the squares-of-distances idea powers the 68-95-99.7 rule.
• Coefficient of Variation — divides $\sigma$ by $\bar{x}$ for a unitless spread.
• Covariance and Correlation — the same arrows-and-products idea, but for two variables at once.
• Least Squares Regression — "minimise the sum of squared arrows," exactly Step 5's tiles.

Concept Map